Professional Documents
Culture Documents
Another Haskell Tutorial
Another Haskell Tutorial
Copyright (c) Hal Daume III, 2002-2006. The preprint version of this tutorial is
intended to be free to the entire Haskell community, so we grant permission to copy
and distribute it for any purpose, provided that it is reproduced in its entirety, including this notice. Modified versions may not be distributed without prior concent
of the author, and must still maintain a copy of this notice. The author retains the
right to change or modify this copyright at any time, as well as to make the book no
longer free of charge.
What is Haskell?
Haskell is called a lazy, pure functional programming language. It is called lazy because expressions which are not needed to determine the answer to a problem are not
evaluated. The opposize of lazy is strict, which is the evaluation strategry of most
common programming languages (C, C++, Java, even ML). A strict language is one in
which every expression is evaluated, whether the result of its computation is important
or not. (This is probably not entirely true as optimizing compilers for strict languages
often do whats called dead code elmination this removes unused expressions from
the program.) It is called pure because it does not allow side effects (A side effect
is something that affects the state of the world. For instance, a function that prints
something to the screen is said to be side-effecting, as is a function which affects the
value of a global variable.) of course, a programming language without side effects
would be horribly useless; Haskell uses a system of monads to isolate all impure computations from the rest of the program and perform them in the safe way (see Chapter 9
for a discussion of monads proper or Chapter 5 for how to do input/output in a pure
language).
Haskell is called a functional language because the evaluation of a program is
equivalent to evaluating a function in the pure mathematical sense. This also differs
from standard languages (like C and Java) which evaluate a sequence of statements,
one after the other (this is termed an imperative langauge).
i
ii
iii
It is intended to be a stable language in sense the implementors are committed to supporting Haskell 98 exactly as specified, for the foreseeable
future.
The original Haskell Report covered only the language, together with a
standard library called the Prelude. By the time Haskell 98 was stabilised,
it had become clear that many programs need access to a larger set of library functions (notably concerning input/output and simple interaction
with the operating system). If these program were to be portable, a set of
libraries would have to be standardised too. A separate effort was therefore begun by a distinct (but overlapping) committee to fix the Haskell 98
Libraries.
iv
Target Audience
There have been many books and tutorials written about Haskell; for a (nearly) complete list, visit the http://haskell.org/bookshelf (Haskell Bookshelf) at the
Haskell homepage. A brief survey of the tutorials available yields:
A Gentle Introduction to Haskell is an introduction to Haskell, given that the
reader is familiar with functional programming en large.
Haskell Companion is a short reference of common concepts and definitions.
Online Haskell Course is a short course (in German) for beginning with Haskell.
Two Dozen Short Lessons in Haskell is the draft of an excellent textbook that
emphasizes user involvement.
Haskell Tutorial is based on a course given at the 3rd International Summer
School on Advanced Functional Programming.
Haskell for Miranda Programmers assumes knowledge of the language Miranda.
PLEAC-Haskell is a tutorial in the style of the Perl Cookbook.
Though all of these tutorials is excellent, they are on their own incomplete: The
Gentle Introduction is far too advanced for beginning Haskellers and the others tend
to end too early, or not cover everything. Haskell is full of pitfalls for new programmers
and experienced non-functional programmers alike, as can be witnessed by reading
through the archives of the Haskell mailing list.
It became clear that there is a strong need for a tutorial which is introductory in the
sense that it does not assume knowledge of functional programming, but which is advanced in the sense that it does assume some background in programming. Moreover,
none of the known tutorials introduce input/output and iteractivity soon enough (Paul
Hudaks book is an exception in that it does introduce IO by page 35, though the focus
and aim of that book and this tutorial are very different). This tutorial is not for beginning programmers; some experience and knowledge of programming and computers is
assumed (though the appendix does contain some background information).
The Haskell language underwent a standardization process and the result is called
Haskell 98. The majority of this book will cover the Haskell 98 standard. Any deviations from the standard will be noted (for instance, many compilers offer certain
extensions to the standard which are useful; some of these may be discussed).
The goals of this tutorial are:
to be practical above all else
to provide a comprehensive, free introduction to the Haskell language
to point out common pitfalls and their solutions
to provide a good sense of how Haskell can be used in the real world
Acknowledgements
It would be inappropriate not to give credit also to the original designers of Haskell.
Those are: Arvind, Lennart Augustsson, Dave Barton, Brian Boutel, Warren Burton,
Jon Fairbairn, Joseph Fasel, Andy Gordon, Maria Guzman, Kevin Hammond, Ralf
Hinze, Paul Hudak, John Hughes, Thomas Johnsson, Mark Jones, Dick Kieburtz, John
Launchbury, Erik Meijer, Rishiyur Nikhil, John Peterson, Simon Peyton Jones, Mike
Reeve, Alastair Reid, Colin Runciman, Philip Wadler, David Wise, Jonathan Young.
Finally, I would like to specifically thank Simon Peyton Jones, Simon Marlow,
John Hughes, Alastair Reid, Koen Classen, Manuel Chakravarty, Sigbjorn Finne and
Sven Panne, all of whom have made my life learning Haskell all the more enjoyable by
always being supportive. There were doubtless others who helped and are not listed,
but these are those who come to mind.
Also thanks to the many people who have reported bugs in the first edition.
- Hal Daume III
vi
Contents
1
Introduction
Getting Started
2.1 Hugs . . . . . . . . . . . . . . . .
2.1.1 Where to get it . . . . . .
2.1.2 Installation procedures . .
2.1.3 How to run it . . . . . . .
2.1.4 Program options . . . . .
2.1.5 How to get help . . . . . .
2.2 Glasgow Haskell Compiler . . . .
2.2.1 Where to get it . . . . . .
2.2.2 Installation procedures . .
2.2.3 How to run the compiler .
2.2.4 How to run the interpreter
2.2.5 Program options . . . . .
2.2.6 How to get help . . . . . .
2.3 NHC . . . . . . . . . . . . . . . .
2.3.1 Where to get it . . . . . .
2.3.2 Installation procedures . .
2.3.3 How to run it . . . . . . .
2.3.4 Program options . . . . .
2.3.5 How to get help . . . . . .
2.4 Editors . . . . . . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
5
5
6
6
6
7
7
7
7
8
8
8
9
9
9
9
9
9
9
9
9
Language Basics
3.1 Arithmetic . . . . . . . . . . .
3.2 Pairs, Triples and More . . . .
3.3 Lists . . . . . . . . . . . . . .
3.3.1 Strings . . . . . . . .
3.3.2 Simple List Functions
3.4 Source Code Files . . . . . . .
3.5 Functions . . . . . . . . . . .
3.5.1 Let Bindings . . . . .
3.5.2 Infix . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
11
13
14
15
17
18
20
22
27
28
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
vii
CONTENTS
viii
3.6
3.7
3.8
4
Comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Recursion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Interactivity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
28
29
31
Type Basics
4.1 Simple Types . . . . . . . . . . .
4.2 Polymorphic Types . . . . . . . .
4.3 Type Classes . . . . . . . . . . .
4.3.1 Motivation . . . . . . . .
4.3.2 Equality Testing . . . . .
4.3.3 The Num Class . . . . . .
4.3.4 The Show Class . . . . . .
4.4 Function Types . . . . . . . . . .
4.4.1 Lambda Calculus . . . . .
4.4.2 Higher-Order Types . . .
4.4.3 That Pesky IO Type . . . .
4.4.4 Explicit Type Declarations
4.4.5 Functional Arguments . .
4.5 Data Types . . . . . . . . . . . .
4.5.1 Pairs . . . . . . . . . . . .
4.5.2 Multiple Constructors . .
4.5.3 Recursive Datatypes . . .
4.5.4 Binary Trees . . . . . . .
4.5.5 Enumerated Sets . . . . .
4.5.6 The Unit type . . . . . . .
4.6 Continuation Passing Style . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
37
37
39
40
40
41
41
41
42
42
42
44
45
46
47
47
49
51
51
52
53
53
Basic Input/Output
5.1 The RealWorld Solution
5.2 Actions . . . . . . . . .
5.3 The IO Library . . . . .
5.4 A File Reading Program
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
57
57
58
62
64
Modules
6.1 Exports . . . . . . . . . . .
6.2 Imports . . . . . . . . . . .
6.3 Hierarchical Imports . . . .
6.4 Literate Versus Non-Literate
6.4.1 Bird-scripts . . . . .
6.4.2 LaTeX-scripts . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
67
67
69
70
71
71
72
Advanced Features
7.1 Sections and Infix Operators
7.2 Local Declarations . . . . .
7.3 Partial Application . . . . .
7.4 Pattern Matching . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
73
73
74
76
80
.
.
.
.
CONTENTS
ix
7.5
7.6
Guards . . . . . . . . . . . . . .
Instance Declarations . . . . . .
7.6.1 The Eq Class . . . . . .
7.6.2 The Show Class . . . .
7.6.3 Other Important Classes
7.6.4 Class Contexts . . . . .
7.6.5 Deriving Classes . . . .
7.7 Datatypes Revisited . . . . . . .
7.7.1 Named Fields . . . . . .
7.8 More Lists . . . . . . . . . . . .
7.8.1 Standard List Functions
7.8.2 List Comprehensions . .
7.9 Arrays . . . . . . . . . . . . . .
7.10 Finite Maps . . . . . . . . . . .
7.11 Layout . . . . . . . . . . . . . .
7.12 The Final Word on Lists . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
83
85
85
86
87
89
89
90
90
92
92
94
96
97
99
99
Advanced Types
8.1 Type Synonyms . . .
8.2 Newtypes . . . . . .
8.3 Datatypes . . . . . .
8.3.1 Strict Fields .
8.4 Classes . . . . . . .
8.4.1 Pong . . . .
8.4.2 Computations
8.5 Instances . . . . . .
8.6 Kinds . . . . . . . .
8.7 Class Hierarchies . .
8.8 Default . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
103
103
104
105
105
108
108
109
113
115
117
117
Monads
9.1 Do Notation . . . . . . . . . . .
9.2 Definition . . . . . . . . . . . .
9.3 A Simple State Monad . . . . .
9.4 Common Monads . . . . . . . .
9.5 Monadic Combinators . . . . .
9.6 MonadPlus . . . . . . . . . . .
9.7 Monad Transformers . . . . . .
9.8 Parsing Monads . . . . . . . . .
9.8.1 A Simple Parsing Monad
9.8.2 Parsec . . . . . . . . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
119
120
122
124
130
134
137
139
144
144
151
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
CONTENTS
10 Advanced Techniques
10.1 Exceptions . . . . . .
10.2 Mutable Arrays . . .
10.3 Mutable References .
10.4 The ST Monad . . .
10.5 Concurrency . . . . .
10.6 Regular Expressions
10.7 Dynamic Types . . .
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
157
157
157
157
157
157
157
157
159
161
C Solutions To Exercises
163
CONTENTS
Chapter 1
Introduction
This tutorial contains a whole host of example code, all of which should have been
included in its distribution. If not, please refer to the links off of the Haskell web site
(haskell.org) to get it. This book is formatted to make example code stand out
from the rest of the text.
Code will look like this.
Occasionally, we will refer to interaction betwen you and the operating system
and/or the interactive shell (more on this in Section 2).
Interaction will look like this.
Strewn throughout the tutorial, we will often make additional notes to something
written. These are often for making comparisons to other programming languages or
adding helpful information.
NOTE Notes will appear like this.
If were covering a difficult or confusing topic and there is something you should
watch out for, we will place a warning.
WARNING Warnings will appear like this.
Finally, we will sometimes make reference to built-in functions (so-called Preludefunctions). This will look something like this:
map :: (a > b) > [a] > [b]
Within the body text, Haskell keywords will appear like this: where, identifiers as
map, types as String and classes as Eq.
3
CHAPTER 1. INTRODUCTION
Chapter 2
Getting Started
There are three well known Haskell system: Hugs, GHC and NHC. Hugs is exclusively
an interpreter, meaning that you cannot compile stand-alone programs with it, but can
test and debug programs in an interactive environment. GHC is both an interpreter (like
Hugs) and a compiler which will produce stand-alone programs. NHC is exclusively a
compiler. Which you use is entirely up to you. Ive tried to make a list of some of the
differences in the following list but of course this is far from exhaustive:
Hugs - very fast; implements almost all of Haskell 98 (the standard) and most extensions; built-in support for module browsing; cannot create stand-alones; written
in C; works on almost every platform; build in graphics library.
GHC - interactive environment is slower than Hugs, but allows function definitions
in the environment (in Hugs you have to put them in a file); implements all of
Haskell 98 and extensions; good support for interfacing with other languages; in
a sense the de facto standard.
NHC - less used and no interactive environment, but produces smaller and often faster
executables than does GHC; supports Haskell 98 and some extensions.
I, personally, have all of them installed and use them for different purposes. I tend
to use GHC to compile (primarily because Im most familiar with it) and the Hugs
interactive environment, since it is much faster. As such, this is what I would suggest.
However, that is a fair amount to download an install, so if you had to go with just one,
Id get GHC, since it contains both a compiler and interactive environment.
Following is a descrition of how to download and install each of this as of the time
this tutorial was written. It may have changed see http://haskell.org (the
Haskell website) for up-to-date information.
2.1 Hugs
Hugs supports almost all of the Haskell 98 standard (it lacks some of the libraries),
as well as a number of advanced/experimental extensions, including: multi-parameter
5
type classes, extensible records, rank-2 polymorphism, existentials, scoped type variables, and restricted type synonyms.
2.1.1
Where to get it
2.1.2
Installation procedures
Once youve downloaded Hugs, installation differs depending on your platform, however, installation for Hugs is more of less identical to installation for any program on
your platform.
For Windows when you click on the msi file to download, simply choose Run This
Program and the installation will begin automatically. From there, just follow
the on-screen instructions.
For RPMs use whatever RPM installation program you know best.
For source first gunzip the file, then untar it. Presumably if youre using a system
which isnt otherwise supported, you know enough about your system to be able
to run configure scripts and make things by hand.
2.1.3
How to run it
On Unix machines, the Hugs interpreter is usually started with a command line of the
form: hugs [option file] ...
On Windows , Hugs may be started by selecting it from the start menu or by double
clicking on a file with the .hs or .lhs extension. (This manual assumes that Hugs has
already been successfully installed on your system.)
Hugs uses options to set system parameters. These options are distinguished by a
leading + or - and are used to customize the behaviour of the interpreter. When Hugs
starts, the interpreter performs the following tasks:
Options in the environment are processed. The variable HUGSFLAGS holds
these options. On Windows 95/NT, the registry is also queried for Hugs option
settings.
Command line options are processed.
Internal data structures are initialized. In particular, the heap is initialized, and
its size is fixed at this point; if you want to run the interpreter with a heap size
other than the default, then this must be specified using options on the command
line, in the environment or in the registry.
The prelude file is loaded. The interpreter will look for the prelude file on the
path specified by the -P option. If the prelude, located in the file Prelude.hs,
cannot be found in one of the path directories or in the current directory, then
Hugs will terminate; Hugs will not run without the prelude file.
Program files specified on the command line are loaded. The effect of a command hugs f1 ... fn is the same as starting up Hugs with the hugs command and
then typing :load f1 ... fn. In particular, the interpreter will not terminate if a
problem occurs while it is trying to load one of the specified files, but it will
abort the attempted load command.
The environment variables and command line options used by Hugs are described
in the following sections.
2.1.4
Program options
To list all of the options would take too much space. The most important option at this
point is +98 or -98. When you start hugs with +98 it is in Haskell 98 mode,
which turns off all extensions. When you start in -98, you are in Hugs mode and all
extensions are turned on. If youve downloaded someone elses code and youre having
trouble loading it, first make sure you have the 98 flag set properly.
Further information on the Hugs options is in the manual:
http://cvs.haskell.org/Hugs/pages/hugsman/started.html ().
2.1.5
To get Hugs specific help, go to the Hugs web page. To get general Haskell help, go to
the Haskell web page.
2.2.1
Where to get it
2.2.2
Installation procedures
Once youve downloaded GHC, installation differs depending on your platform; however, installation for GHC is more of less identical to installation for any program on
your platform.
For Windows when you click on the msi file to download, simply choose Run This
Program and the installation will begin automatically. From there, just follow
the on-screen instructions.
For RPMs use whatever RPM installation program you know best.
For source first gunzip the file, then untar it. Presumably if youre using a system
which isnt otherwise supported, you know enough about your system to be able
to run configure scripts and make things by hand.
For a more detailed description of the installation procedure, look at the GHC users
manual under Installing GHC.
2.2.3
Running the compiler is fairly easy. Assuming that you have a program written with a
main function in a file called Main.hs, you can compile it simply by writing:
% ghc --make Main.hs -o main
The make option tells GHC that this is a program and not just a library and you
want to build it and all modules it depends on. Main.hs stipulates the name of the
file to compile; and the -o main means that you want to put the output in a file called
main.
2.2.4
GHCi is invoked with the command ghci or ghc interactive. One or more modules
or filenames can also be specified on the command line; this instructs GHCi to load the
specified modules or filenames (and all the modules they depend on), just as if you had
said :load modules at the GHCi prompt.
2.3. NHC
2.2.5
Program options
To list all of the options would take too much space. The most important option at
this point is -fglasgow-exts. When you start GHCi without -fglasgow-exts it is
in Haskell 98 mode, which turns off all extensions. When you start with -fglasgowexts, all extensions are turned on. If youve downloaded someone elses code and
youre having trouble loading it, first make sure you have this flag set properly.
Further information on the GHC and GHCi options are in the manual off of the
GHC web page.
2.2.6
To get GHC(i) specific help, go to the GHC web page. To get general Haskell help, go
to the Haskell web page.
2.3 NHC
About NHC. . .
2.3.1
Where to get it
2.3.2
Installation procedures
2.3.3
How to run it
2.3.4
Program options
2.3.5
2.4 Editors
With good text editor, programming is fun. Of course, you can get along with simplistic
editor capable of just cut-n-paste, but good editor is capable of doing most of the chores
for you, letting you concentrate on what you are writing. With respect to programming
in Haskell, good text editor should have as much as possible of the following features:
Syntax highlighting for source files
Indentation of source files
Interaction with Haskell interpreter (be it Hugs or GHCi)
Computer-aided code navigation
Code completion
10
Chapter 3
Language Basics
In this chapter we present the basic concepts of Haskell. In addition to familiarizing you
with the interactive environments and showing you how to compile a basic program,
we introduce the basic syntax of Haskell, which will probably be quite alien if you are
used to languages like C and Java.
However, before we talk about specifics of the language, we need to establish some
general properties of Haskell. Most importantly, Haskell is a lazy language, which
means that no computation takes place unless it is forced to take place when the result
of that computation is used.
This means, for instance, that you can define infinitely large data structures, provided that you never use the entire structure. For instance, using imperative-esque
psuedo-code, we could create an infinite list containing the number 1 in each position
by doing something like:
lazy
List makeList()
{
List current = new List();
current.value = 1;
current.next = makeList();
return current;
}
By looking at this code, we can see what its trying to do: it creates a new list, sets
its value to 1 and then recursively calls itself to make the rest of the list. Of course, if
you actually wrote this code and called it, the program would never terminate, because
makeList would keep calling itself ad infinitum.
This is because we assume this imperative-esque language is strict, the opposite of
lazy. Strict languages are often referred to as call by value, while lazy languages are
referred to as call by name. In the above psuedo-code, when we run makeList
on the fifth line, we attempt to get a value out of it. This leads to an infinite loop.
The equivalent code in Haskell is:
11
strict
12
makeList = 1 : makeList
case-sensitive
values
types
side effects
pure
This program reads: were defining something called makeList (this is what goes
on the left-hand side of the equals sign). On the right-hand side, we give the definition
of makeList. In Haskell, the colon operator is used to create lists (well talk more
about this soon). This right-hand side says that the value of makeList is the element
1 stuck on to the beginning of the value of makeList.
However, since Haskell is lazy (or call by name), we do not actually attempt to
evaluate what makeList is at this point: we simply remember that if ever in the future
we need the second element of makeList, we need to just look at makeList.
Now, if you attempt to write makeList to a file, print it to the screen, or calculate
the sum of its elements, the operation wont terminate because it would have to evaluate
an infinitely long list. However, if you simply use a finite portion of the list (say the
first 10 elements), the fact that the list is infinitely long doesnt matter. If you only use
the first 10 elements, only the first 10 elements are ever calculated. This is laziness.
Second, Haskell is case-sensitive. Many languages are, but Haskell actually uses
case to give meaning. Haskell distinguishes between values (for instance, numbers:
1, 2, 3, . . . ); strings: abc, hello, . . . ; characters: a, b, , . . . ; even functions:
for instance, the function squares a value, or the square-root function); and types (the
categories to which values belong).
By itself, this is not unusual. Most languages have some system of types. What is
unusual is that Haskell requires that the names given to functions and values begin with
a lower-case letter and that the names given to types begin with an upper-case letter.
The moral is: if your otherwise correct program wont compile, be sure you havent
named your function Foo, or something else beginning with a capital letter.
Being a functional language, Haskell eschews side effects. A side effect is essentially something that happens in the course of executing a function that is not related to
the output produced by that function.
For instance, in a language like C or Java, you are able to modify global variables
from within a function. This is a side effect because the modification of this global
variable is not related to the output produced by the function. Furthermore, modifying
the state of the real world is considered a side effect: printing something to the screen,
reading a file, etc., are all side effecting operations.
Functions that do not have side effects are called pure. An easy test for whether or
not a function is pure is to ask yourself a simple question: Given the same arguments,
will this function always produce the same result?.
All of this means that if youre used to writing code in an imperative language (like
C or Java), youre going to have to start thinking differently. Most importantly, if you
have a value x, you must not think of x as a register, a memory location or anything
else of that nature. x is simply a name, just as Hal is my name. You cannot arbitrarily
decide to store a different person in my name any more than you can arbitrarily decide
to store a different value in x. This means that code that might look like the following
C code is invalid (and has no counterpart) in Haskell:
int x = 5;
3.1. ARITHMETIC
13
x = x + 1;
A call like x = x + 1 is called destructive update because we are destroying
whatever was in x before and replacing it with a new value. Destructive update does
not exist in Haskell.
By not allowing destructive updates (or any other such side effecting operations),
Haskell code is very easy to comprehend. That is, when we define a function f, and
call that function with a particular argument a in the beginning of a program, and then,
at the end of the program, again call f with the same argument a, we know we will
get out the same result. This is because we know that a cannot have changed and
because we know that f only depends on a (for instance, it didnt increment a global
counter). This property is called referential transparency and basically states that if
two functions f and g produce the same values for the same arguments, then we may
replace f with g (and vice-versa).
destructive update
referential transparency
NOTE There is no agreed-upon exact definition of referential transparency. The definition given above is the one I like best. They all carry
the same interpretation; the differences lie in how they are formalized.
3.1 Arithmetic
Lets begin our foray into Haskell with simple arithmetic. Start up your favorite interactive shell (Hugs or GHCi; see Chapter 2 for installation instructions). The shell will
output to the screen a few lines talking about itself and what its doing and then should
finish with the cursor on a line reading:
Prelude>
From here, you can begin to evaluate expressions. An expression is basically something that has a value. For instance, the number 5 is an expression (its value is 5). Values can be built up from other values; for instance, 5 + 6 is an expression (its value is
11). In fact, most simple arithmetic operations are supported by Haskell, including plus
(+), minus (-), times (*), divided-by (/), exponentiation () and square-root (sqrt).
You can experiment with these by asking the interactive shell to evaluate expressions
and to give you their value. In this way, a Haskell shell can be used as a powerful
calculator. Try some of the following:
Prelude> 5*4+3
23
Prelude> 55-2
3123
Prelude> sqrt 2
1.4142135623730951
expressions
14
Prelude> 5*(4+3)
35
operator precedence
We can see that, in addition to the standard arithmetic operations, Haskell also
allows grouping by parentheses, hence the difference between the values of 5*4+3 and
5*(4+3). The reason for this is that the understood grouping of the first expression is
(5*4)+3, due to operator precedence.
Also note that parentheses arent required around function arguments. For instance,
we simply wrote sqrt 2, not sqrt(2), as would be required in most other languages. You could write it with the parentheses, but in Haskell, since function application is so common, parentheses arent required.
WARNING Even though parentheses are not always needed, sometimes it is better to leave them in anyway; other people will probably
have to read your code, and if extra parentheses make the intent of the
code clearer, use them.
Now try entering 25000. Does it work?
NOTE If youre familiar with programming in other languages, you
may find it odd that sqrt 2 comes back with a decimal point (i.e., is a
floating point number) even though the argument to the function seems to
be an integer. This interchangability of numeric types is due to Haskells
system of type classes and will be discussed in detail in Section 4.3).
Exercises
Exercise 3.1 Weve seen that multiplication binds more tightly than division. Can you
think of a way to determine whether function application binds more or less tightly
than multiplication?
ogeneous
3.3. LISTS
15
Here, we have a pair of integers, 5 and 3. In Haskell, the first element of a pair
need not have the same type as the second element: that is, pairs are allowed to be
heterogeneous. For instance, you can have a pair of an integer with a string. This
contrasts with lists, which must be made up of elements of all the same type (we will
discuss lists further in Section 3.3).
There are two predefined functions that allow you to extract the first and second
elements of a pair. They are, respectively, fst and snd. You can see how they work
below:
Prelude> fst (5, "hello")
5
Prelude> snd (5, "hello")
"hello"
In addition to pairs, you can define triples, quadruples etc. To define a triple and a
quadruple, respectively, we write:
Prelude> (1,2,3)
(1,2,3)
Prelude> (1,2,3,4)
(1,2,3,4)
And so on. In general, pairs, triples, and so on are called tuples and can store fixed
amounts of heterogeneous data.
NOTE The functions fst and snd wont work on anything longer
than a pair; if you try to use them on a larger tuple, you will get a message
stating that there was a type error. The meaning of this error message will
be explained in Chapter 4.
Exercises
Exercise 3.2 Use a combination of fst and snd to extract the character out of the tuple
((1,a),"foo").
3.3 Lists
The primary limitation of tuples is that they hold only a fixed number of elements:
pairs hold two, triples hold three, and so on. A data structure that can hold an arbitrary
number of elements is a list. Lists are assembled in a very similar fashion to tuples,
except that they use square brackets instead of parentheses. We can define a list like:
tuples
16
Prelude> [1,2]
[1,2]
Prelude> [1,2,3]
[1,2,3]
cons operator
Lists dont need to have any elements. The empty list is simply [].
Unlike tuples, we can very easily add an element on to the beginning of the list
using the colon operator. The colon is called the cons operator; the process of adding
an element is called consing. The etymology of this is that we are constructing a
new list from an element and an old list. We can see the cons operator in action in the
following examples:
Prelude> 0:[1,2]
[0,1,2]
Prelude> 5:[1,2,3,4]
[5,1,2,3,4]
We can actually build any list by using the cons operator (the colon) and the empty
list:
Prelude> 5:1:2:3:4:[]
[5,1,2,3,4]
syntactic sugar
In fact, the [5,1,2,3,4] syntax is syntactic sugar for the expression using the
explicit cons operators and empty list. If we write something using the [5,1,2,3,4]
notation, the compiler simply translates it to the expression using (:) and [].
homogenous
One further difference between lists and tuples is that, while tuples are heterogeneous, lists must be homogenous. This means that you cannot have a list that holds
both integers and strings. If you try to, a type error will be reported.
Of course, lists dont have to just contain integers or strings; they can also contain
tuples or even other lists. Tuples, similarly, can contain lists and other tuples. Try some
of the following:
Prelude> [(1,1),(2,4),(3,9),(4,16)]
[(1,1),(2,4),(3,9),(4,16)]
Prelude> ([1,2,3,4],[5,6,7])
([1,2,3,4],[5,6,7])
3.3. LISTS
17
There are two basic list functions: head and tail. The head function returns
the first element of a (non-empty) list, and the tail function returns all but the first
element of a (non-empty) list.
To get the length of a list, you use the length function:
Prelude> length [1,2,3,4,10]
5
Prelude> head [1,2,3,4,10]
1
Prelude> length (tail [1,2,3,4,10])
4
3.3.1
Strings
In Haskell, a String is simply a list of Chars. So, we can create the string Hello as:
Prelude> H:e:l:l:o:[]
"Hello"
Lists (and, of course, strings) can be concatenated using the ++ operator:
Prelude> "Hello " ++ "World"
"Hello World"
Additionally, non-string values can be converted to strings using the show function, and strings can be converted to non-string values using the read function. Of
course, if you try to read a value thats malformed, an error will be reported (note that
this is a run-time error, not a compile-time error):
Prelude> "Five squared is " ++ show (5*5)
"Five squared is 25"
Prelude> read "5" + 3
8
Prelude> read "Hello" + 3
Program error: Prelude.read: no parse
Above, the exact error message is implementation dependent. However, the interpreter has inferred that youre trying to add three to something. This means that when
we execute read "Hello", we expect to be returned a number. However, "Hello"
cannot be parsed as a number, so an error is reported.
18
3.3.2
Much of the computation in Haskell programs is done by processing lists. There are
three primary list-processing functions: map, filter and foldr (also foldl).
The map function takes as arguments a list of values and a function that should be
applied to each of the values. For instance, there is a built-in function Char.toUpper
that takes as input a Char and produces a Char that is the upper-case version of the
original argument. So, to covert an entire string (which is simply a list of characters)
to upper case, we can map the toUpper function across the entire list:
Prelude> map Char.toUpper "Hello World"
"HELLO WORLD"
WARNING Hugs users: Hugs doesnt like qualified names like
Char.toUpper. In Hugs, simply use toUpper.
When you map across a list, the length of the list never changes only the individual values in the list change.
To remove elements from the list, you can use the filter function. This function
allows you to remove certain elements from a list depending on their value, but not on
their context. For instance, the function Char.isLower tells you whether a given
character is lower case. We can filter out all non-lowercase characters using this:
Prelude> filter Char.isLower "Hello World"
"elloorld"
The function foldr takes a little more getting used to. foldr takes three arguments: a function, an initial value and a list. The best way to think about foldr is
that it replaces occurences of the list cons operator (:) with the function parameter and
replaces the empty list constructor ([]) with the initial value. Thus, if we have a list:
3 : 8 : 12 : 5 : []
and we apply foldr (+) 0 to it, we get:
3 + 8 + 12 + 5 + 0
which sums the list. We can test this:
Prelude> foldr (+) 0 [3,8,12,5]
28
We can perform the same sort of operation to calculate the product of all the elements on a list:
Prelude> foldr (*) 1 [4,8,5]
160
3.3. LISTS
19
We said earlier that folding is like replacing (:) with a particular function and ([])
with an initial element. This raises a question as to what happens when the function
isnt associative (a function () is associative if a (b c) = (a b) c). When we write
4 8 5 1, we need to specify where to put the parentheses. Namely, do we mean
((4 8) 5) 1 or 4 (8 ((5 1))? foldr assumes the function is right-associative (i.e.,
the correct bracketing is the latter). Thus, when we use it on a non-associtive function
(like minus), we can see the effect:
Prelude> foldr (-) 1 [4,8,5]
0
The exact derivation of this looks something like:
==>
==>
==>
==>
==>
==>
==>
The foldl function goes the other way and effectively produces the opposite
bracketing. foldl looks the same when applied, so we could have done summing
just as well with foldl:
Prelude> foldl (+) 0 [3,8,12,5]
28
However, we get different results when using the non-associative function minus:
Prelude> foldl (-) 1 [4,8,5]
-16
This is because foldl uses the opposite bracketing. The way it accomplishes this
is essentially by going all the way down the list, taking the last element and combining
it with the initial value via the provided function. It then takes the second-to-last element in the list and combines it to this new value. It does so until there is no more list
left.
The derivation here proceeds in the opposite fashion:
==>
==>
==>
foldl
foldl
foldl
foldl
(-)
(-)
(-)
(-)
1 [4,8,5]
(1 - 4) [8,5]
((1 - 4) - 8) [5]
(((1 - 4) - 8) - 5) []
20
==>
==>
==>
==>
((1 - 4) - 8) - 5
((-3) - 8) - 5
(-11) - 5
-16
Note that once the foldl goes away, the parenthesization is exactly the opposite
of the foldr.
NOTE foldl is often more efficient than foldr for reasons that
we will discuss in Section 7.8. However, foldr can work on infinite
lists, while foldl cannot. This is because before foldl does anything, it has to go to the end of the list. On the other hand, foldr
starts producing output immediately. For instance, foldr (:) []
[1,2,3,4,5] simply returns the same list. Even if the list were infinite, it would produce output. A similar function using foldl would
fail to produce any output.
If this discussion of the folding functions is still somewhat unclear, thats okay.
Well discuss them further in Section 7.8.
Exercises
Exercise 3.3 Use map to convert a string into a list of booleans, each element in the
new list representing whether or not the original element was a lower-case character.
That is, it should take the string aBCde and return [True,False,False,True,True].
Exercise 3.4 Use the functions mentioned in this section (you will need two of them)
to compute the number of lower-case letters in a string. For instance, on aBCde it
should return 3.
Exercise 3.5 Weve seen how to calculate sums and products using folding functions.
Given that the function max returns the maximum of two numbers, write a function
using a fold that will return the maximum value in a list (and zero if the list is empty).
So, when applied to [5,10,2,8,1] it will return 10. Assume that the values in the list are
always 0. Explain to yourself why it works.
Exercise 3.6 Write a function that takes a list of pairs of length at least 2 and returns the first component of the second element in the list. So, when provided with
[(5,b),(1,c),(6,a)], it will return 1.
21
module Test
where
x = 5
y = (6, "Hello")
z = x * fst y
This is a very simple program written in Haskell. It defines a module called
Test (in general module names should match file names; see Section 6 for more on
this). In this module, there are three definitions: x, y and z. Once youve written
and saved this file, in the directory in which you saved it, load this in your favorite
interpreter, by executing either of the following:
% hugs Test.hs
% ghci Test.hs
This will start Hugs or GHCi, respectively, and load the file. Alternatively, if you
already have one of them loaded, you can use the :load command (or just :l) to
load a module, as:
Prelude> :l Test.hs
...
Test>
Between the first and last line, the interpreter will print various data to explain what
it is doing. If any errors appear, you probably mistyped something in the file; double
check and then try again.
Youll notice that where it used to say Prelude it now says Test. That means
that Test is the current module. Youre probably thinking that Prelude must also be
a module. Exactly correct. The Prelude module (usually simply referred to as the
Prelude) is always loaded and contains the standard definitions (for instance, the (:)
operator for lists, or (+) or (*), fst, snd and so on).
Now that weve loaded Test, we can use things that were defined in it. For example:
Test> x
5
Test> y
(6,"Hello")
Test> z
30
Perfect, just as we expected!
One final issue regarding how to compile programs to stand-alone executables remains. In order for a program to be an executable, it must have the module name
the Prelude
22
Main and must contain a function called main. So, if you go in to Test.hs and rename it to Main (change the line that reads module Test to module Main), we
simply need to add a main function. Try this:
main = putStrLn "Hello World"
Now, save the file, and compile it (refer back to Section 2 for information on how
to do this for your compiler). For example, in GHC, you would say:
% ghc --make Test.hs -o test
3.5 Functions
Now that weve seen how to write code in a file, we can start writing functions. As you
might have expected, functions are central to Haskell, as it is a functional language.
This means that the evaluation of a program is simply the evaluation of a function.
We can write a simple function to square a number and enter it into our Test.hs
file. We might define this as follows:
square x = x * x
In this function definition, we say that were defining a function square that takes
one argument (aka parameter), which we call x. We then say that the value of square
x is equal to x * x.
Haskell also supports standard conditional expressions. For instance, we could
define a function that returns 1 if its argument is less than 0; 0 if its argument is 0;
and 1 if its argument is greater than 0 (this is called the signum function):
3.5. FUNCTIONS
23
signum x =
if x < 0
then -1
else if x > 0
then 1
else 0
You can experiment with this as:
Test>
1
Test>
0
Test>
-1
Test>
-1
signum 5
signum 0
signum (5-10)
signum (-1)
Note that the parenthesis around -1 in the last example are required; if missing,
the system will think you are trying to subtract the value 1 from the value signum,
which is illtyped.
The if/then/else construct in Haskell is very similar to that of most other programming languages; however, you must have both a then and an else clause. It evaluates
the condition (in this case x < 0 and, if this evaluates to True, it evaluates the then
condition; if the condition evaluated to False, it evaluates the else condition).
You can test this program by editing the file and loading it back into your interpreter.
If Test is already the current module, instead of typing :l Test.hs again, you can
simply type :reload or just :r to reload the current file. This is usually much faster.
Haskell, like many other languages, also supports case constructions. These are
used when there are multiple values that you want to check against (case expressions
are actually quite a bit more powerful than this see Section 7.4 for all of then details).
Suppose we wanted to define a function that had a value of 1 if its argument were
0; a value of 5 if its argument were 1; a value of 2 if its argument were 2; and a value of
1 in all other instances. Writing this function using if statements would be long and
very unreadable; so we write it using a case statement as follows (we call this function
f):
f x =
case x
0 ->
1 ->
2 ->
_ ->
of
1
5
2
-1
if/then/else
case/of
24
wildcard
In this program, were defining f to take an argument x and then inspect the value
of x. If it matches 0, the value of f is 1. If it matches 1, the value of f is 5. If it maches
2, then the value of f is 2; and if it hasnt matched anything by that point, the value of
f is 1 (the underscore can be thought of as a wildcard it will match anything) .
The indentation here is important. Haskell uses a system called layout to structure its code (the programming language Python uses a similar system). The layout
system allows you to write code without the explicit semicolons and braces that other
languages like C and Java require.
WARNING Because whitespace matters in Haskell, you need to be
careful about whether you are using tabs or spaces. If you can configure
your editor to never use tabs, thats probably better. If not, make sure
your tabs are always 8 spaces long, or youre likely to run in to problems.
The general rule for layout is that an open-brace is inserted after the keywords
where, let, do and of, and the column position at which the next command appears
is remembered. From then on, a semicolon is inserted before every new line that is
indented the same amount. If a following line is indented less, a close-brace is inserted.
This may sound complicated, but if you follow the general rule of indenting after each
of those keywords, youll never have to remember it (see Section 7.11 for a more
complete discussion of layout).
Some people prefer not to use layout and write the braces and semicolons explicitly.
This is perfectly acceptable. In this style, the above function might look like:
f x = case x of
{ 0 -> 1 ; 1 -> 5 ; 2 -> 2 ; _ -> 1 }
Of course, if you write the braces and semicolons explicitly, youre free to structure
the code as you wish. The following is also equally valid:
f x =
case x of { 0 -> 1 ;
1 -> 5 ; 2 -> 2
; _ -> 1 }
However, structuring your code like this only serves to make it unreadable (in this
case).
Functions can also be defined piece-wise, meaning that you can write one version
of your function for certain parameters and then another version for other parameters.
For instance, the above function f could also be written as:
f
f
f
f
0
1
2
_
=
=
=
=
1
5
2
-1
3.5. FUNCTIONS
25
Here, the order is important. If we had put the last line first, it would have matched
every argument, and f would return -1, regardless of its argument (most compilers
will warn you about this, though, saying something about overlapping patterns). If we
had not included this last line, f would produce an error if anything other than 0, 1 or
2 were applied to it (most compilers will warn you about this, too, saying something
about incomplete patterns). This style of piece-wise definition is very popular and will
be used quite frequently throughout this tutorial. These two definitions of f are actually
equivalent this piece-wise version is translated into the case expression.
More complicated functions can be built from simpler functions using function
composition. Function composition is simply taking the result of the application of one
function and using that as an argument for another. Weve already seen this way back
in arithmetic (Section 3.1), when we wrote 5*4+3. In this, we were evaluating 5 4
and then applying +3 to the result. We can do the same thing with our square and f
functions:
Test>
25
Test>
4
Test>
5
Test>
-1
square (f 1)
square (f 2)
f (square 1)
f (square 2)
function composition
26
Test> (square . f) 2
4
Test> (f . square) 1
5
Test> (f . square) 2
-1
sqrt 2
id "hello"
id 5
fst (5,2)
snd (5,2)
null []
null [1,2,3,4]
head [1,2,3,4]
tail [1,2,3,4]
[1,2,3] ++ [4,5,6]
3.5. FUNCTIONS
27
[1,2,3,4,5,6]
Prelude> [1,2,3] == [1,2,3]
True
Prelude> a /= b
True
Prelude> head []
Program error: {head []}
We can see that applying head to an empty list gives an error (the exact error
message depends on whether youre using GHCi or Hugs the shown error message is
from Hugs).
3.5.1
Let Bindings
Often we wish to provide local declarations for use in our functions. For instance, if
you remember back to your grade school mathematics courses, the following equation
is used
to find the roots (zeros) of a polynomial of the form ax2 + bx + c = 0: x =
(b b2 4ac)/2a. We could write the following function to compute the two
values of x:
roots a b c =
((-b + sqrt(b*b - 4*a*c)) / (2*a),
(-b - sqrt(b*b - 4*a*c)) / (2*a))
To remedy this problem, Haskell allows for local bindings. That is, we can create
values inside of a function that only that function can see. For instance, we could
create a local binding for sqrt(b*b-4*a*c) and call it, say, det and then use that
in both places where sqrt(b*b - 4*a*c) occurred. We can do this using a let/in
declaration:
roots a b c =
let det = sqrt (b*b - 4*a*c)
in ((-b + det) / (2*a),
(-b - det) / (2*a))
In fact, you can provide multiple declarations inside a let. Just make sure theyre
indented the same amount, or you will have layout problems:
roots a b c =
let det = sqrt (b*b - 4*a*c)
twice_a = 2*a
in ((-b + det) / twice_a,
(-b - det) / twice_a)
28
3.5.2
Infix
Infix functions are ones that are composed of symbols, rather than letters. For instance,
(+), (*), (++) are all infix functions. You can use them in non-infix mode by
enclosing them in parentheses. Hence, the two following expressions are the same:
Prelude> 5 + 10
15
Prelude> (+) 5 10
15
Similarly, non-infix functions (like map) can be made infix by enclosing them in
backquotes (the ticks on the tilde key on American keyboards):
Prelude> map Char.toUpper "Hello World"
"HELLO WORLD"
Prelude> Char.toUpper map "Hello World"
"HELLO WORLD"
3.6 Comments
There are two types of comments in Haskell: line comments and block comments.
Line comments begin with the token -- and extend until the end of the line. Block
comments begin with {- and extend to a corresponding -}. Block comments can be
nested.
NOTE The -- in Haskell corresponds to // in C++ or Java, and {and -} correspond to /* and */.
Comments are used to explain your program in English and are completely ignored
by compilers and interpreters. For example:
module Test2
where
main =
putStrLn "Hello World"
-- write a string
-- to the screen
3.7. RECURSION
29
3.7 Recursion
In imperative languages like C and Java, the most basic control structure is a loop (like
a for loop). However, for loops dont make much sense in Haskell because they require
destructive update (the index variable is constantly being updated). Instead, Haskell
uses recursion.
A function is recursive if it calls itself (see Appendix B for more). Recursive functions exist also in C and Java but are used less than they are in functional languages.
The prototypical recursive function is the factorial function. In an imperative language,
you might write this as something like:
int factorial(int n) {
int fact = 1;
for (int i=2; i <= n; i++)
fact = fact * i;
return fact;
}
While this code fragment will successfully compute factorials for positive integers,
it somehow ignores the basic definition of factorial, usually given as:
1
n=1
n! =
n (n 1)! otherwise
This definition itself is exactly a recursive definition: namely the value of n! depends on the value of (n 1)!. If you think of ! as a function, then it is calling itself.
We can translate this definition almost verbatim into Haskell code:
factorial 1 = 1
factorial n = n * factorial (n-1)
This is likely the simplest recursive function youll ever see, but it is correct.
30
a
a ab1
b=1
otherwise
3.8. INTERACTIVITY
31
Similarly, we can consider the filter function. Again, the base case is the empty
list, and the recursive case is a cons list. However, this time, were choosing whether
to keep an element, depending on whether or not a particular predicate holds. We can
define the filter function as:
my_filter p [] = []
my_filter p (x:xs) =
if p x
then x : my_filter p xs
else my_filter p xs
In this code, when presented with an empty list, we simply return an empty list.
This is because filter cannot add elements; it can only remove them.
When presented with a list of the form (x:xs), we need to decide whether or not
to keep the value x. To do this, we use an if statement and the predicate p. If p x is
true, then we return a list that begins with x followed by the result of filtering the tail
of the list. If p x is false, then we exclude x and return the result of filtering the tail of
the list.
We can also define map and both fold functions using explicit recursion. See the
exercises for the definition of map and Chapter 7 for the folds.
Exercises
Exercise 3.7 The fibonacci sequence is defined by:
1
n = 1 or n = 2
Fn =
Fn2 + Fn1 otherwise
Write a recursive function fib that takes a positive integer n as a parameter and
calculates Fn .
Exercise 3.8 Define a recursive function mult that takes two positive integers a and
b and returns a*b, but only uses addition (i.e., no fair just using multiplication). Begin
by making a mathematical definition in the style of the previous exercise and the rest of
this section.
Exercise 3.9 Define a recursive function my map that behaves identically to the standard function map.
3.8 Interactivity
If you are familiar with books on other (imperative) languages, you might be wondering why you havent seen many of the standard programs written in tutorials of other
languages (like ones that ask the user for his name and then says Hi to him by name).
The reason for this is simple: Being a pure functional language, it is not entirely clear
how one should handle operations like user input.
32
monads
do notation
After all, suppose you have a function that reads a string from the keyboard. If
you call this function twice, and the user types something the first time and something
else the second time, then you no longer have a function, since it would return two
different values. The solution to this was found in the depths of category theory, a
branch of formal mathematics: monads. Were not yet ready to talk about monads
formally, but for now, think of them simply as a convenient way to express operations
like input/output. Well discuss them in this context much more in Chapter 5 and then
discuss monads for monads sake in Chapter 9.
Suppose we want to write a function thats interactive. The way to do this is to
use the do keyword. This allows us to specify the order of operations (remember that
normally, since Haskell is a lazy language, the order in which operations are evaluated
in it is unspecified). So, to write a simple program that asks a user for his name and
then address him directly, enter the following code into Name.hs:
module Main
where
import IO
main = do
hSetBuffering stdin LineBuffering
putStrLn "Please enter your name: "
name <- getLine
putStrLn ("Hello, " ++ name ++ ", how are you?")
3.8. INTERACTIVITY
33
functions. On the seventh line, we start with do, telling Haskell that were executing a
sequence of commands.
The first command is hSetBuffering stdin LineBuffering, which you
should probably ignore for now (incidentally, this is only required by GHC in Hugs
you can get by without it). The necessity for this is because, when GHC reads input, it
expects to read it in rather large blocks. A typical persons name is nowhere near large
enough to fill this block. Thus, when we try to read from stdin, it waits until its
gotten a whole block. We want to get rid of this, so we tell it to use LineBuffering
instead of block buffering.
The next command is putStrLn, which prints a string to the screen. On the
ninth line, we say name <- getLine. This would normally be written name =
getLine, but using the arrow instead of the equal sign shows that getLine isnt
a real function and can return different values. This command means run the action
getLine, and store the results in name.
The last line constructs a string using what we read in the previous line and then
prints it to the screen.
Another example of a function that isnt really a function would be one that returns
a random value. In this context, a function that does this is called randomRIO. Using this, we can write a guess the number program. Enter the following code into
Guess.hs:
module Main
where
import IO
import Random
main = do
hSetBuffering stdin LineBuffering
num <- randomRIO (1::Int, 100)
putStrLn "Im thinking of a number between 1 and 100"
doGuessing num
doGuessing num = do
putStrLn "Enter your guess:"
guess <- getLine
let guessNum = read guess
if guessNum < num
then do putStrLn "Too low!"
doGuessing num
else if read guess > num
then do putStrLn "Too high!"
doGuessing num
else do putStrLn "You Win!"
Lets examine this code. On the fifth line we write import Random to tell the
34
compiler that were going to be using some random functions (these arent built into
the Prelude). In the first line of main, we ask for a random number in the range
(1, 100). We need to write ::Int to tell the compiler that were using integers here
not floating point numbers or other numbers. Well talk more about this in Section 4.
On the next line, we tell the user whats going on, and then, on the last line of main,
we tell the compiler to execute the command doGuessing.
The doGuessing function takes the number the user is trying to guess as an
argument. First, it asks the user to guess and then accepts their guess (which is a
String) from the keyboard. The if statement checks first to see if their guess is too
low. However, since guess is a string, and num is an integer, we first need to convert
guess to an integer by reading it. Since read guess is a plain, pure function
(and not an IO action), we dont need to use the <- notation (in fact, we cannot); we
simply bind the value to guessNum. Note that while were in do notation, we dont
need ins for lets.
If they guessed too low, we inform them and then start doGuessing over again.
If they didnt guess too low, we check to see if they guessed too high. If they did,
we tell them and start doGuessing again. Otherwise, they didnt guess too low and
they didnt guess too high, so they must have gotten it correct. We tell them that they
won and exit. The fact that we exit is implicit in the fact that there are no commands
following this. We dont need an explicit return () statement.
You can either compile this code or load it into your interpreter, and you will get
something like:
Main> main
Im thinking of a number between 1 and 100
Enter your guess:
50
Too low!
Enter your guess:
75
Too low!
Enter your guess:
85
Too high!
Enter your guess:
80
Too high!
Enter your guess:
78
Too low!
Enter your guess:
79
You Win!
The recursive action that we just saw doesnt actually return a value that we use in
any way. In the case when it does, the obvious way to write the command is actually
3.8. INTERACTIVITY
35
incorrect. Here, we will give the incorrect version, explain why it is wrong, then give
the correct version.
Lets say were writing a simple program that repeatedly asks the user to type in a
few words. If at any point the user enters the empty word (i.e., he just hits enter without
typing anything), the program prints out everything hes typed up until that point and
then exits. The primary function (actually, an action) in this program is one that asks
the user for a word, checks to see if its empty, and then either continues or ends. The
incorrect formulation of this might look something like:
askForWords = do
putStrLn "Please enter a word:"
word <- getLine
if word == ""
then return []
else return (word : askForWords)
Before reading ahead, see if you can figure out what is wrong with the above code.
The error is on the last line, specifically with the term word : askForWords.
Remeber that when using (:), we are making a list out of an element (in this case
word) and another list (in this case, askForWords). However, askForWords is
not a list; it is an action that, when run, will produce a list. That means that before we
can attach anything to the front, we need to run the action and take the result. In this
case, we want to do something like:
askForWords = do
putStrLn "Please enter a word:"
word <- getLine
if word == ""
then return []
else do
rest <- askForWords
return (word : rest)
Here, we first run askForWords, take the result and store it in the variable rest.
Then, we return the list created from word and rest.
By now, you should have a good understanding of how to write simple functions,
compile them, test functions and programs in the interactive environment, and manipulate lists.
Exercises
Exercise 3.10 Write a program that will repeatedly ask the user for numbers until she
types in zero, at which point it will tell her the sum of all the numbers, the product of
all the numbers, and, for each number, its factorial. For instance, a session might look
like:
36
0 to stop):
0 to stop):
0 to stop):
0 to stop):
Hint: write an IO action that reads a number and, if its zero, returns the empty list. If
its not zero, it recurses itself and then makes a list out of the number it just read and
the result of the recursive call.
Chapter 4
Type Basics
Haskell uses a system of static type checking. This means that every expression in
Haskell is assigned a type. For instance a would have type Char, for character.
Then, if you have a function which expects an argument of a certain type and you give
it the wrong type, a compile-time error will be generated (that is, you will not be able
to compile the program). This vastly reduces the number of bugs that can creep into
your program.
Furthermore, Haskell uses a system of type inference. This means that you dont
even need to specify the type of expressions. For comparison, in C, when you define
a variable, you need to specify its type (for instance, int, char, etc.). In Haskell, you
neednt do this the type will be inferred from context.
NOTE If you want, you certainly are allowed to explicitely specify
the type of an expression; this often helps debugging. In fact, it is sometimes considered good style to explicitly specify the types of outermost
functions.
Both Hugs and GHCi allow you to apply type inference to an expression to find its
type. This is done by using the :t command. For instance, start up your favorite shell
and try the following:
Prelude> :t c
c :: Char
This tells us that the expression c has type Char (the double colon :: is used
throughout Haskell to specify types).
38
strings), and others. We have already seen an expression of type Char; lets examime
one of type String:
Prelude> :t "Hello"
"Hello" :: String
You can also enter more complicated expressions, for instance, a test of equality:
Prelude> :t a == b
a == b :: Bool
You should note than even though this expression is false, it still has a type, namely
the type Bool.
NOTE Bool is short for Boolean (usually pronounced boo-lee-uhn,
though Ive heard boo-leen once or twice) and has two possible values:
True and False.
You can observe the process of type checking and type inference by trying to get
the shell to give you the type of an ill-typed expression. For instance, the equality
operator requires that the type of both of its arguments are of the same type. We can
see that Char and String are of different types by trying to compare a character to a
string:
Prelude> :t a ==
ERROR - Type error
*** Expression
*** Term
*** Type
*** Does not match
"a"
in application
: a == "a"
: a
: Char
: [Char]
The first line of the error (the line containing Expression) tells us the expression
in which the type error occured. The second line tells us which part of this expression
is ill-typed. The third line tells us the inferred type of this term and the fourth line
tells us what it needs to have matched. In this case, it says that type type Char doesnt
match the type [Char] (a list a characters a string in Haskell is represented as a list of
characters).
As mentioned before, you can explicitely specify the type of an expression using
the :: operator. For instance, instead of a in the previous example, we could have
written (a::String). In this case, this has no effect since theres only one possible
interpretation of a. However, consider the case of numbers. You can try:
Prelude> :t 5 :: Int
5 :: Int
Prelude> :t 5 :: Double
5 :: Double
39
Here, we can see that the number 5 can be instantiated as either an Int our a Double.
What if we dont specify the type?
Prelude> :t 5
5 :: Num a => a
Not quite what you expected? What this means, briefly, is that if some type a is
an instance of the Num class, then type type of the expression 5 can be of type a.
If that made no sense, thats okay for now. In Section 4.3 we talk extensively about
type classes (which is what this is). The way to read this, though, is to say a being an
instance of Num implies a.
Exercises
Exercise 4.1 Figure out for yourself, and then verify the types of the following expressions, if they have a type. Also note if the expression is a type error:
1. h:e:l:l:o:[]
2. [5,a]
3. (5,a)
4. (5::Int) + 10
5. (5::Int) + (10::Double)
40
Here, GHCi has made explicit the universal quantification of the type values. That
is, it is saying that for all types a and b, fst is a function from (a, b) to a.
Exercises
Exercise 4.2 Figure out for yourself, and then verify the types of the following expressions, if they have a type. Also note if the expression is a type error:
1. snd
2. head
3. null
4. head .
tail
5. head .
head
4.3.1
Motivation
In many languages (C++, Java, etc.), there exists a system of overloading. That is,
a function can be written that takes parameters of differing types. For instance, the
cannonical example is the equality function. If we want to compare two integers, we
should use an integer comparison; if we want to compare two floating point numbers,
we should use a floating point comparison; if we want to compare two characters,
we should use a character comparison. In general, if we want to compare two things
which have type , we want to use an -compare. We call a type variable since it is
a variable whose value is a type.
NOTE In general, type variables will be written using the first part of
the Greek alphabet: , , , , . . . .
Unfortunately, this presents some problems for static type checking, since the type
checker doesnt know which types a certain operation (for instance, equality testing)
will be defined for. There are as many solutions to this problem as there are statically
typed languages (perhaps a slight exageration, but not so much so). The one chosen
in Haskell is the system of type classes. Whether this is the correct solution or the
best solution of course depends on your application domain. It is, however, the one
we have, so you should learn to love it.
4.3.2
41
Equality Testing
4.3.3
In addition to overloading operators like ==, Haskell has overloaded numeric constants
(i.e., 1, 2, 3, etc.). This was done so that when you type in a number like 5, the
compiler is free to say 5 is an integer or floating point number as it sees fit. It defines
the Num class to contain all of these numbers and certain minimal operations over
them (addition, for instance). The basic numeric types (Int, Double) are defined to be
instances of Num.
We have only skimmed the surface of the power (and complexity) of type classes
here. There will be much more discussion of them in Section 8.4, but we need some
more background before we can get there. Before we do that, we need to talk a little
more about functions.
4.3.4
Another of the standard classes in Haskell is the Show class. Types which are members of the Show class have functions which convert values of that type to a string.
This function is called show. For instance show applied to the integer 5 is the string
5; show applied to the character a is the three-character string a (the first and
last characters are apostrophes). show applied to a string simply puts quotes around it.
You can test this in the interpreter:
Prelude>
"5"
Prelude>
"a"
Prelude>
"\"Hello
show 5
show a
show "Hello World"
World\""
NOTE The reason the backslashes appear in the last line is because
the interior quotes are escaped, meaning that they are part of the string,
not part of the interpreter printing the value. The actual string doesnt
contain the backslashes.
Some types are note instances of Show; functions for example. If you try to show a
function (like sqrt), the compiler or interpreter will give you some cryptic error mes-
42
4.4.1
Lambda Calculus
The name Lambda Calculus, while perhaps daunting, describes a fairly simple system for representing functions. The way we would write a squaring function in lambda
calculus is: x.xx, which means that we take a value, which we will call x (thats what
x. means) and then multiply it by itself. The is called lambda abstraction. In
general, lambdas can only have one parameter. If we want to write a function that takes
two numbers, doubles the first and adds it to the second, we would write: xy.2x+y.
When we apply a value to a lambda expression, we remove the outermost and replace
every occurrence of the lambda variable with the value. For instance, if we evaluate
(x.x x)5, we remove the lambda and replace every occurrence of x with 5, yielding
(5 5) which is 25.
In fact, Haskell is largely based on an extension of the lambda calculus, and these
two expressions can be written directly in Haskell (we simply replace the with a
backslash and the . with an arrow; also we dont need to repeat the lambdas; and, of
course, in Haskell we have to give them names if were defining functions):
square = \x -> x*x
f = \x y -> 2*x + y
You can also evaluate lambda expressions in your interactive shell:
Prelude> (\x -> x*x) 5
25
Prelude> (\x y -> 2*x + y) 5 4
14
We can see in the second example that we need to give the lambda abstraction two
arguments, one corresponding to x and the other corresponding to y.
4.4.2
Higher-Order Types
Higher-Order Types is the name given to functions. The type given to functions mimicks the lambda calculus representation of the functions. For instance, the definition of
square gives x.x x. To get the type of this, we first ask ourselves what the type of x
43
is. Say we decide x is an Int. Then, we notice that the function square takes an Int
and produces a value x*x. We know that when we multiply two Ints together, we get
another Int, so the type of the results of square is also an Int. Thus, we say the type
of square is Int Int.
We can apply a similar analysis to the function f above. The value of this function
(remember, functions are values) is something which takes a value x and given that
value, produces a new value, which takes a value y and produces 2*x+y. For instance,
if we take f and apply only one number to it, we get (xy.2x + y)5 which becomes
our new value y.2(5) + y, where all occurances of x have been replaced with the
applied value, 5.
So we know that f takes an Int and produces a value of some type, of which were
not sure. But we know the type of this value is the type of y.2(5) + y. We apply the above analysis and find out that this expression has type Int Int. Thus, f
takes an Int and produces something which has type Int Int. So the type of f is
Int (Int Int).
NOTE The parentheses are not necessary; in function types, if you
have it is assume that is grouped. If you want the
other way, with grouped, you need to put parentheses around
them.
This isnt entirely accurate. As we saw before, numbers like 5 arent really of type
Int, they are of type Num a a.
We can easily find the type of Prelude functions using :t as before:
Prelude> :t head
head :: [a] -> a
Prelude> :t tail
tail :: [a] -> [a]
Prelude> :t null
null :: [a] -> Bool
Prelude> :t fst
fst :: (a,b) -> a
Prelude> :t snd
snd :: (a,b) -> b
We read this as: head is a function that takes a list containing values of type a
and gives back a value of type a; tail takes a list of as and gives back another list
of as; null takes a list of as and gives back a boolean; fst takes a pair of type
(a,b) and gives back something of type a, and so on.
NOTE Saying that the type of fst is (a, b) a does not necessarily
mean that it simply gives back the first element; it only means that it
gives back something with the same type as the first element.
44
We can also get the type of operators like + and * and ++ and :; however, in order
to do this we need to put them in parentheses. In general, any function which is used
infix (meaning in the middle of two arguments rather than before them) must be put in
parentheses when getting its type.
Prelude> :t (+)
(+) :: Num a => a -> a -> a
Prelude> :t (*)
(*) :: Num a => a -> a -> a
Prelude> :t (++)
(++) :: [a] -> [a] -> [a]
Prelude> :t (:)
(:) :: a -> [a] -> [a]
The types of + and * are the same, and mean that + is a function which, for some
type a which is an instance of Num, takes a value of type a and produces another
function which takes a value of type a and produces a value of type a. In short hand,
we might say that + takes two values of type a and produces a value of type a, but this
is less precise.
The type of ++ means, in shorthand, that, for a given type a, ++ takes two lists of
as and produces a new list of as. Similarly, : takes a value of type a and another value
of type [a] (list of as) and produces another value of type [a].
4.4.3
You might be tempted to try getting the type of a function like putStrLn:
Prelude>
putStrLn
Prelude>
readFile
:t
::
:t
::
putStrLn
String -> IO ()
readFile
FilePath -> IO String
What in the world is that IO thing? Its basically Haskells way of representing
that these functions arent really functions. Theyre called IO Actions (hence the
IO). The immediate question which arises is: okay, so how do I get rid of the IO.
In brief, you cant directly remove it. That is, you cannot write a function with type
IO String String. The only way to use things with an IO type is to combine them
with other functions using (for example), the do notation.
For example, if youre reading a file using readFile, presumably you want to do
something with the string it returns (otherwise, why would you read the file in the first
place). Suppose you have a function f which takes a String and produces an Int. You
cant directly apply f to the result of readFile since the input to f is String and the
output of readFile is IOString and these dont match. However, you can combine
these as:
45
main = do
s <- readFile "somefile"
let i = f s
putStrLn (show i)
Here, we use the arrow convention to get the string out of the IO action and then
apply f to the string (called s). We then, for example, print i to the screen. Note that
the let here doesnt have a corresponding in. This is because we are in a do block.
Also note that we dont write i <- f s because f is just a normal function, not an
IO action.
4.4.4
46
function definition since it knows you will only apply square to Ints, so it may be
able to generate faster code.
If you have extensions turned on (-98 in Hugs or -fglasgow-exts in GHC(i)),
you can also add a type signature to expressions and not just functions. For instance,
you could write:
square (x :: Int) = x*x
which tells the compiler that x is an Int; however, it leaves the compiler alone
to infer the type of the rest of the expression. What is the type of square in this
example? Make your guess then you can check it either by entering this code into a file
and loading it into your interpreter or by asking for the type of the expression:
Prelude> :t (\(x :: Int) -> x*x)
since this lambda abstraction is equivalent to the above function declaration.
4.4.5
Functional Arguments
In Section 3.3 we saw examples of functions taking other functions as arguments. For
instance, map took a function to apply to each element in a list, filter took a function that told it which elements of a list to keep, and foldl took a function which told
it how to combine list elements together. As with every other function in Haskell, these
are well-typed.
Lets first think about the map function. Its job is to take a list of elements and
produce another list of elements. These two lists dont necessarily have to have the
same types of elements. So map will take a value of type [a] and produce a value of
type [b]. How does it do this? It uses the user-supplied function to convert. In order
to convert an a to a b, this function must have type a b. Thus, the type of map is
(a b) [a] [b], which you can verify in your interpreter with :t.
We can apply the same sort of analysis to filter and discern that it has type
(a Bool) [a] [a]. As we presented the foldl function, you might be tempted
to give it type (a a a) a [a] a, meaning that you take a function which
combines two as into another one, an initial value of type a, a list of as to produce a final
value of type a. In fact, foldl has a more general type: (a b a) a [b] a.
So it takes a function which turn an a and a b into an a, an initial value of type a and a
list of bs. It produces an a.
To see this, we can write a function count which counts how many members of a
list satisfy a given constraint. You can of course you filter and length to do this,
but we will also do it using foldr:
module Count
where
import Char
47
Exercises
Exercise 4.3 Figure out for yourself, and then verify the types of the following expressions, if they have a type. Also note if the expression is a type error:
1. \x -> [x]
2. \x y z -> (x,y:z:[])
3. \x -> x + 5
4. \x -> "hello, world"
5. \x -> x a
6. \x -> x x
7. \x -> x + x
4.5.1
Pairs
For instance, a definition of a pair of elements (much like the standard, build-in pair
type) could be:
data Pair a b = Pair a b
48
Lets walk through this code one word at a time. First we say data meaning that
were defining a datatype. We then give the name of the datatype, in this case, Pair.
The a and b that follow Pair are type parameters, just like the a is the type of
the function map. So up until this point, weve said that were going to define a data
structure called Pair which is parameterized over two types, a and b.
After the equals sign, we specify the constructors of this data type. In this case,
there is a single constructor, Pair (this doesnt necessarily have to have the same
name as the type, but in this case it seems to make more sense). After this pair, we
again write a b, which means that in order to construct a Pair we need two values,
one of type a and one of type b.
This definition introduces a function, Pair :: a -> b -> Pair a b that
you can use to construct Pairs. If you enter this code into a file and load it, you can
see how these are constructed:
Datatypes> :t Pair
Pair :: a -> b -> Pair a b
Datatypes> :t Pair a
Pair a :: a -> Pair Char a
Datatypes> :t Pair a "Hello"
:t Pair a "Hello"
Pair a "Hello" :: Pair Char [Char]
So, by giving Pair two values, we have completely constructed a value of type
Pair. We can write functions involving pairs as:
pairFst (Pair x y) = x
pairSnd (Pair x y) = y
In this, weve used the pattern matching capabilities of Haskell to look at a pair
an extract values from it. In the definition of pairFst we take an entire Pair and
extract the first element; similarly for pairSnd. Well discuss pattern matching in
much more detail in Section 7.4.
Exercises
Exercise 4.4 Write a data type declaration for Triple, a type which contains three
elements, all of different types. Write functions tripleFst, tripleSnd and tripleThr
to extract respectively the first, second and third elements.
Exercise 4.5 Write a datatype Quadruple which holds four elements. However, the
first two elements must be the same type and the last two elements must be the same
type. Write a function firstTwo which returns a list containing the first two elements
and a function lastTwo which returns a list containing the last two elements. Write
type signatures for these functions
4.5.2
49
Multiple Constructors
We have seen an example of the data type with one constructor: Pair. It is also
possible (and extremely useful) to have multiple constructors.
Let us consider a simple function which searches through a list for an element
satisfying a given predicate and then returns the first element satisfying that predicate.
What should we do if none of the elements in the list satisfy the predicate? A few
options are listed below:
Raise an error
Loop indefinitely
Write a check function
Return the first element
...
Raising an error is certainly an option (see Section 10.1 to see how to do this).
The problem is that it is difficult/impossible to recover from such errors. Looping
indefinitely is possible, but not terribly useful. We could write a sister function which
checks to see if the list contains an element satisfying a predicate and leave it up to the
user to always use this function first. We could return the first element, but this is very
ad-hoc and difficult to remember.
The fact that there is no basic option to solve this problem simply means we have to
think about it a little more. What are we trying to do? Were trying to write a function
which might succeed and might not. Furthermore, if it does succeed, it returns some
sort of value. Lets write a datatype:
data Maybe a = Nothing
| Just a
This is one of the most common datatypes in Haskell and is defined in the Prelude.
Here, were saying that there are two possible ways to create something of type
Maybe a. The first is to use the nullary constructor Nothing, which takes no arguments (this is what nullary means). The second is to use the constructor Just,
together with a value of type a.
The Maybe type is useful in all sorts of circumstances. For instance, suppose
we want to write a function (like head) which returns the first element of a given
list. However, we dont want the program to die if the given list is empty. We can
accomplish this with a function like:
firstElement :: [a] -> Maybe a
firstElement []
= Nothing
firstElement (x:xs) = Just x
50
The type signature here says that firstElement takes a list of as and produces
something with type Maybe a. In the first line of code, we match against the empty
list []. If this match succeeds (i.e., the list is, in fact, empty), we return Nothing. If
the first match fails, then we try to match against x:xs which must succeed. In this
case, we return Just x.
For our findElement function, we represent failure by the value Nothing and
success with value a by Just a. Our function might look something like this:
findElement :: (a -> Bool) -> [a] -> Maybe a
findElement p [] = Nothing
findElement p (x:xs) =
if p x then Just x
else findElement p xs
The first line here gives the type of the function. In this case, our first argument
is the predicate (and takes an element of type a and returns True if and only if the
element satisfies the predicate); the second argument is a list of as. Our return value
is maybe an a. That is, if the function succeeds, we will return Just a and if not,
Nothing.
Another useful datatype is the Either type, defined as:
data Either a b = Left a
| Right b
Exercises
Exercise 4.6 Write a datatype Tuple which can hold one, two, three or four elements,
depending on the constructor (that is, there should be four constructors, one for each
number of arguments). Also provide functions tuple1 through tuple4 which take a
tuple and return Just the value in that position, or Nothing if the number is invalid
(i.e., you ask for the tuple4 on a tuple holding only two elements).
Exercise 4.7 Based on our definition of Tuple from the previous exercise, write a
function which takes a Tuple and returns either the value (if its a one-tuple), a
Haskell-pair (i.e., (a,5)) if its a two-tuple, a Haskell-triple if its a three-tuple
or a Haskell-quadruple if its a four-tuple. You will need to use the Either type to
represent this.
4.5.3
51
Recursive Datatypes
We can also define recursive datatypes. These are datatypes whose definitions are
based on themselves. For instance, we could define a list datatype as:
data List a = Nil
| Cons a (List a)
In this definition, we have defined what it means to be of type List a. We say that a
list is either empty (Nil) or its the Cons of a value of type a and another value of type
List a. This is almost identical to the actual definition of the list datatype in Haskell,
except that uses special syntax where [] corresponds to Nil and : corresponds to
Cons. We can write our own length function for our lists as:
listLength Nil = 0
listLength (Cons x xs) = 1 + listLength xs
This function is slightly more complicated and uses recursion to calculate the
length of a List. The first line says that the length of an empty list (a Nil) is 0.
This much is obvious. The second line tells us how to calculate the length of a nonempty list. A non-empty list must be of the form Cons x xs for some values of x
and xs. We know that xs is another list and we know that whatever the length of the
current list is, its the length of its tail (the value of xs) plus one (to account for x).
Thus, we apply the listLength function to xs and add one to the result. This gives
us the length of the entire list.
Exercises
Exercise 4.8 Write functions listHead, listTail, listFoldl and listFoldr
which are equivalent to their Prelude twins, but function on our List datatype. Dont
worry about exceptional conditions on the first two.
4.5.4
Binary Trees
We can define datatypes that are more complicated that lists. Suppose we want to
define a structure that looks like a binary tree. A binary tree is a structure that has a
single root node; each node in the tree is either a leaf or a branch. If its a leaf, it
holds a value; if its a branch, it holds a value and a left child and a right child. Each of
these children is another node. We can define such a data type as:
data BinaryTree a
= Leaf a
| Branch (BinaryTree a) a (BinaryTree a)
In this datatype declaration we say that a BinaryTree of as is either a Leaf
which holds an a, or its a branch with a left child (which is a BinaryTree of as), a
52
node value (which is an a), and a right child (which is also a BinaryTree of as). It is
simple to modify the listLength function so that instead of calculating the length
of lists, it calculates the number of nodes in a BinaryTree. Can you figure out how?
We can call this function treeSize. The solution is given below:
treeSize (Leaf x) = 1
treeSize (Branch left x right) =
1 + treeSize left + treeSize right
Here, we say that the size of a leaf is 1 and the size of a branch is the size of its left
child, plus the size of its right child, plus one.
Exercises
Exercise 4.9 Write a function elements which returns the elements in a BinaryTree
in a bottom-up, left-to-right manner (i.e., the first element returned in the left-most leaf,
followed by its parents value, followed by the other childs value, and so on). The result type should be a normal Haskell list.
Exercise 4.10 Write a fold function for BinaryTrees and rewrite elements in
terms of it (call the new one elements2).
4.5.5
Enumerated Sets
You can also use datatypes to define things like enumerated sets, for instance, a type
which can only have a constrained number of values. We could define a color type:
data Color
= Red
| Orange
| Yellow
| Green
| Blue
| Purple
| White
| Black
This would be sufficient to deal with simple colors. Suppose we were using this to
write a drawing program, we could then write a function to convert between a Color
and a RGB triple. We can write a colorToRGB function, as:
colorToRGB
colorToRGB
colorToRGB
colorToRGB
colorToRGB
Red
Orange
Yellow
Green
Blue
=
=
=
=
=
(255,0,0)
(255,128,0)
(255,255,0)
(0,255,0)
(0,0,255)
53
-- R G B components
4.5.6
A final useful datatype defined in Haskell (from the Prelude) is the unit type. Its
definition is:
data () = ()
The only true value of this type is (). This is essentially the same as a void type in
a langauge like C or Java and will be useful when we talk about IO in Chapter 5.
Well dwell much more on data types in Sections 7.4 and 8.3.
54
parsing problem. So is the task of taking an English sentence and creating a parse tree
(though the latter is quite a bit harder).
Suppose were parsing something like C or Java where functions take arguments
in parentheses. But for simplicity, assume they are not separated by commas. That
is, a function call looks like myFunction(x y z). We want to convert this into
something like a pair containing first the string myFunction and then a list with three
string elements: x, y and z.
The general approach to solving this would be to write a function which parses
function calls like this one. First it would look for an identifier (myFunction), then
for an open parenthesis, then for zero or more identifiers, then for a close parenthesis.
One way to do this would be to have two functions:
parseFunction
::
[Token] -> Maybe ((String, [String]), [Token])
parseIdentifier ::
[Token] -> Maybe (String, [Token])
The idea would be that if we call parseFunction, if it doesnt return Nothing,
then it returns the pair described earlier, together with whatever is left after parsing the
function. Similarly, parseIdentifier will parse one of the arguments. If it returns
Nothing, then its not an argument; if it returns Just something, then that something
is the argument paired with the rest of the tokens.
What the parseFunction function would do is to parse an identifier. If this
fails, it fails itself. Otherwise, it continues and tries to parse a open parenthesis. If that
succeeds, it repeatedly calls parseIdentifier until that fails. It then tries to parse
a close parenthesis. If that succeeds, then its done. Otherwise, it fails.
There is, however, another way to think about this problem. The advantage to this
solution is that functions no longer need to return the remaining tokens (which tends to
get ugly). Instead of the above, we write functions:
parseFunction
::
[Token] -> ((String, [String]) -> [Token] -> a) ->
([Token] -> a) -> a
parseIdentifier ::
[Token] -> (String -> [Token] -> a) ->
([Token] -> a) -> a
Lets consider parseIdentifier. This takes three arguments: a list of tokens
and two continuations. The first continuation is what to do when you succeed. The
second continuation is what to do if you fail. What parseIdentifier does, then,
is try to read an identifier. If this succeeds, it calls the first continuation with that
identifier and the remaining tokens as arguments. If reading the identifier fails, it calls
the second continuation with all the tokens.
55
56
The only difference between these calls to cfold is whether we call the continuation before or after constructing the list. As it turns out, this slight difference changes
the behavior for being like foldr to being like foldl. We can evaluate both of these
calls as follows (let f be the folding function):
==>
==>
==>
==>
==>
==>
==>
==>
==>
==>
==>
==>
==>
==>
==>
==>
==>
==>
==>
==>
==>
==>
Exercises
Exercise 4.11 Test whether the CPS-style fold mimicks either of foldr and foldl.
If not, where is the difference?
Exercise 4.12 Write map and filter using continuation passing style.
Chapter 5
Basic Input/Output
As we mentioned earlier, it is difficult to think of a good, clean way to integrate operations like input/output into a pure functional language. Before we give the solution,
lets take a step back and think about the difficulties inherent in such a task.
Any IO library should provide a host of functions, containing (at a minimum) operations like:
print a string to the screen
read a string from a keyboard
write data to a file
read data from a file
There are two issues here. Lets first consider the initial two examples and think
about what their types should be. Certainly the first operation (I hesitate to call it a
function) should take a String argument and produce something, but what should it
produce? It could produce a unit (), since there is essentially no return value from
printing a string. The second operation, similarly, should return a String, but it doesnt
seem to require an argument.
We want both of these operations to be functions, but they are by definition not
functions. The item that reads a string from the keyboard cannot be a function, as it
will not return the same String every time. And if the first function simply returns ()
every time, there should be no problem with replacing it with a function f
= (),
due to referential transparency. But clearly this does not have the desired effect.
58
5.2 Actions
The breakthrough for solving this problem came when Phil Wadler realized that monads would be a good way to think about IO computations. In fact, monads are able to
express much more than just the simple operations described above; we can use them
to express a variety of constructions like concurrence, exceptions, IO, non-determinism
and much more. Moreover, there is nothing special about them; they can be defined
within Haskell with no special handling from the compiler (though compilers often
choose to optimize monadic operations).
5.2. ACTIONS
59
As pointed out before, we cannot think of things like print a string to the screen or
read data from a file as functions, since they are not (in the pure mathematical sense).
Therefore, we give them another name: actions. Not only do we give them a special
name, we give them a special type. One particularly useful action is putStrLn, which
prints a string to the screen. This action has type:
putStrLn :: String -> IO ()
As expected, putStrLn takes a string argument. What it returns is of type IO
(). This means that this function is actually an action (that is what the IO means).
Furthermore, when this action is evaluated (or run) , the result will have type ().
NOTE Actually, this type means that putStrLn is an action within
the IO monad, but we will gloss over this for now.
You can probably already guess the type of getLine:
getLine :: IO String
This means that getLine is an IO action that, when run, will have type String.
The question immediately arises: how do you run an action?. This is something
that is left up to the compiler. You cannot actually run an action yourself; instead, a
program is, itself, a single action that is run when the compiled program is executed.
Thus, the compiler requires that the main function have type IO (), which means that
it is an IO action that returns nothing. The compiled code then executes this action.
However, while you are not allowed to run actions yourself, you are allowed to
combine actions. In fact, we have already seen one way to do this using the do
notation (how to really do this will be revealed in Chapter 9). Lets consider the original
name program:
main = do
hSetBuffering stdin LineBuffering
putStrLn "Please enter your name: "
name <- getLine
putStrLn ("Hello, " ++ name ++ ", how are you?")
We can consider the do notation as a way to combine a sequence of actions. Moreover, the <- notation is a way to get the value out of an action. So, in this program,
were sequencing four actions: setting buffering, a putStrLn, a getLine and another putStrLn. The putStrLn action has type String IO (), so we provide it a
String, so the fully applied action has type IO (). This is something that we are allowed
to execute.
The getLine action has type IO String, so it is okay to execute it directly. However, in order to get the value out of the action, we write name <- getLine, which
basically means run getLine, and put the results in the variable called name.
60
Normal Haskell constructions like if/then/else and case/of can be used within the
do notation, but you need to be somewhat careful. For instance, in our guess the
number program, we have:
do ...
if (read guess) < num
then do putStrLn "Too low!"
doGuessing num
else if read guess > num
then do putStrLn "Too high!"
doGuessing num
else do putStrLn "You Win!"
If we think about how the if/then/else construction works, it essentially takes three
arguments: the condition, the then branch, and the else branch. The condition
needs to have type Bool, and the two branches can have any type, provided that they
have the same type. The type of the entire if/then/else construction is then the type of
the two branches.
In the outermost comparison, we have (read guess) < num as the condition.
This clearly has the correct type. Lets just consider the then branch. The code here
is:
do putStrLn "Too low!"
doGuessing num
Here, we are sequencing two actions: putStrLn and doGuessing. The first
has type IO (), which is fine. The second also has type IO (), which is fine. The type
result of the entire computation is precisely the type of the final computation. Thus, the
type of the then branch is also IO (). A similar argument shows that the type of the
else branch is also IO (). This means the type of the entire if/then/else construction
is IO (), which is just what we want.
NOTE In this code, the last line is else do putStrLn "You
Win!". This is somewhat overly verbose. In fact, else putStrLn
"You Win!" would have been sufficient, since do is only necessary to
sequence actions. Since we have only one action here, it is superfluous.
5.2. ACTIONS
61
Here, since we didnt repeat the do, the compiler doesnt know that the putStrLn
and doGuessing calls are supposed to be sequenced, and the compiler will think
youre trying to call putStrLn with three arguments: the string, the function doGuessing
and the integer num. It will certainly complain (though the error may be somewhat difficult to comprehend at this point).
We can write the same doGuessing function using a case statement. To do this,
we first introduce the Prelude function compare, which takes two values of the same
type (in the Ord class) and returns one of GT, LT, EQ, depending on whether the first
is greater than, less than or equal to the second.
doGuessing num = do
putStrLn "Enter your guess:"
guess <- getLine
case compare (read guess) num of
LT -> do putStrLn "Too low!"
doGuessing num
GT -> do putStrLn "Too high!"
doGuessing num
EQ -> putStrLn "You Win!"
Here, again, the dos after the ->s are necessary on the first two options, because
we are sequencing actions.
If youre used to programming in an imperative language like C or Java, you might
think that return will exit you from the current function. This is not so in Haskell.
In Haskell, return simply takes a normal value (for instance, one of type IO Int) and
makes it into an action that returns the given value (for instance, the value of type Int).
In particular, in an imperative language, you might write this function as:
void doGuessing(int num) {
print "Enter your guess:";
int guess = atoi(readLine());
if (guess == num) {
print "You win!";
return ();
}
// we wont get here if guess == num
if (guess < num) {
print "Too low!";
doGuessing(num);
} else {
print "Too high!";
doGuessing(num);
}
}
62
Here, because we have the return () in the first if match, we expect the code
to exit there (and in mode imperative languages, it does). However, the equivalent code
in Haskell, which might look something like:
doGuessing num = do
putStrLn "Enter your guess:"
guess <- getLine
case compare (read guess) num of
EQ -> do putStrLn "You win!"
return ()
-- we dont expect to get here unless guess == num
if (read guess < num)
then do print "Too low!";
doGuessing
else do print "Too high!";
doGuessing
will not behave as you expect. First of all, if you guess correctly, it will first print You
win!, but it wont exit, and it will check whether guess is less than num. Of course
it is not, so the else branch is taken, and it will print Too high! and then ask you to
guess again.
On the other hand, if you guess incorrectly, it will try to evaluate the case statement
and get either LT or GT as the result of the compare. In either case, it wont have a
pattern that matches, and the program will fail immediately with an exception.
Exercises
Exercise 5.1 Write a program that asks the user for his or her name. If the name is
one of Simon, John or Phil, tell the user that you think Haskell is a great programming
language. If the name is Koen, tell them that you think debugging Haskell is fun (Koen
Classen is one of the people who works on Haskell debugging); otherwise, tell the user
that you dont know who he or she is.
Write two different versions of this program, one using if statements, the other using a
case statement.
= ReadMode
| WriteMode
| AppendMode | ReadWriteMode
openFile
:: Handle -> IO ()
hIsEOF
63
hGetChar
:: Handle -> IO Char
hGetLine
:: Handle -> IO String
hGetContents :: Handle -> IO String
getChar
getLine
getContents
:: IO Char
:: IO String
:: IO String
hPutChar
hPutStr
hPutStrLn
putChar
putStr
putStrLn
:: Char -> IO ()
:: String -> IO ()
:: String -> IO ()
readFile
writeFile
bracket
::
IO a -> (a -> IO b) -> (a -> IO c) -> IO c
NOTE The type FilePath is a type synonym for String. That is,
there is no difference between FilePath and String. So, for instance,
the readFile function takes a String (the file to read) and returns an
action that, when run, produces the contents of that file. See Section 8.1
for more about type synonyms.
Most of these functions are self-explanatory. The openFile and hClose functions open and close a file, respectively, using the IOMode argument as the mode for
opening the file. hIsEOF tests for end-of file. hGetChar and hGetLine read
a character or line (respectively) from a file. hGetContents reads the entire file.
The getChar, getLine and getContents variants read from standard input.
hPutChar prints a character to a file; hPutStr prints a string; and hPutStrLn
prints a string with a newline character at the end. The variants without the h prefix
work on standard output. The readFile and writeFile functions read an entire
file without having to open it first.
The bracket function is used to perform actions safely. Consider a function that
opens a file, writes a character to it, and then closes the file. When writing such a
function, one needs to be careful to ensure that, if there were an error at some point,
the file is still successfully closed. The bracket function makes this easy. It takes
64
three arguments: The first is the action to perform at the beginning. The second is the
action to perform at the end, regardless of whether theres an error or not. The third is
the action to perform in the middle, which might result in an error. For instance, our
character-writing function might look like:
writeChar :: FilePath -> Char -> IO ()
writeChar fp c =
bracket
(openFile fp ReadMode)
hClose
(\h -> hPutChar h c)
This will open the file, write the character and then close the file. However, if
writing the character fails, hClose will still be executed, and the exception will be
reraised afterwards. That way, you dont need to worry too much about catching the
exceptions and about closing all of your handles.
65
doRead filename =
bracket (openFile filename ReadMode) hClose
(\h -> do contents <- hGetContents h
putStrLn "The first 100 chars:"
putStrLn (take 100 contents))
doWrite filename = do
putStrLn "Enter text to go into the file:"
contents <- getLine
bracket (openFile filename WriteMode) hClose
(\h -> hPutStrLn h contents)
What does this program do? First, it issues a short string of instructions and reads
a command. It then performs a case switch on the command and checks first to see if
the first character is a q. If it is, it returns a value of unit type.
NOTE The return function is a function that takes a value of type
a and returns an action of type IO a. Thus, the type of return () is
IO ().
If the first character of the command wasnt a q, the program checks to see if it
was an r followed by some string that is bound to the variable filename. It then
tells you that its reading the file, does the read and runs doLoop again. The check
for w is nearly identical. Otherwise, it matches , the wildcard character, and loops
to doLoop.
The doRead function uses the bracket function to make sure there are no problems reading the file. It opens a file in ReadMode, reads its contents and prints the first
100 characters (the take function takes an integer n and a list and returns the first n
elements of the list).
The doWrite function asks for some text, reads it from the keyboard, and then
writes it to the file specified.
NOTE Both doRead and doWrite could have been made simpler
by using readFile and writeFile, but they were written in the extended fashion to show how the more complex functions are used.
The only major problem with this program is that it will die if you try to read a file
that doesnt already exists or if you specify some bad filename like *\# @. You may
think that the calls to bracket in doRead and doWrite should take care of this,
but they dont. They only catch exceptions within the main body, not within the startup
or shutdown functions (openFile and hClose, in these cases). We would need to
catch exceptions raised by openFile, in order to make this complete. We will do this
when we talk about exceptions in more detail in Section 10.1.
66
Exercises
Exercise 5.2 Write a program that first asks whether the user wants to read from a file,
write to a file or quit. If the user responds quit, the program should exit. If he responds
read, the program should ask him for a file name and print that file to the screen (if the
file doesnt exist, the program may crash). If he responds write, it should ask him for a
file name and then ask him for text to write to the file, with . signaling completion.
All but the . should be written to the file.
For example, running this program might produce:
Do you want to [read] a file, [write] a file
read
Enter a file name to read:
foo
...contents of foo...
Do you want to [read] a file, [write] a file
write
Enter a file name to write:
foo
Enter text (dot on a line by itself to end):
this is some
text for
foo
.
Do you want to [read] a file, [write] a file
read
Enter a file name to read:
foo
this is some
text for
foo
Do you want to [read] a file, [write] a file
read
Enter a file name to read:
foof
Sorry, that file does not exist.
Do you want to [read] a file, [write] a file
blech
I dont understand the command blech.
Do you want to [read] a file, [write] a file
quit
Goodbye!
or [quit]?
or [quit]?
or [quit]?
or [quit]?
or [quit]?
or [quit]?
Chapter 6
Modules
In Haskell, program subcomponents are divided into modules. Each module sits in its
own file and the name of the module should match the name of the file (without the
.hs extension, of course), if you wish to ever use that module in a larger program.
For instance, suppose I am writing a game of poker. I may wish to have a separate
module called Cards to handle the generation of cards, the shuffling and the dealing
functions, and then use this Cards module in my Poker modules. That way, if I
ever go back and want to write a blackjack program, I dont have to rewrite all the code
for the cards; I can simply import the old Cards module.
6.1 Exports
Suppose as suggested we are writing a cards module. I have left out the implementation
details, but suppose the skeleton of our module looks something like this:
module Cards
where
data Card = ...
data Deck = ...
newDeck :: ... -> Deck
newDeck = ...
shuffle :: ... -> Deck -> Deck
shuffle = ...
-- deal deck n deals n cards from deck
deal :: Deck -> Int -> [Card]
deal deck n = dealHelper deck n []
67
CHAPTER 6. MODULES
68
dealHelper = ...
In this code, the function deal calls a helper function dealHelper. The implementation of this helper function is very dependent on the exact data structures you
used for Card and Deck so we dont want other people to be able to call this function.
In order to do this, we create an export list, which we insert just after the module name
declaration:
module Cards ( Card(),
Deck(),
newDeck,
shuffle,
deal
)
where
...
Here, we have specified exactly what functions the module exports, so people who
use this module wont be able to access our dealHelper function. The () after
Card and Deck specify that we are exporting the type but none of the constructors.
For instance if our definition of Card were:
data Card =
data Suit =
|
|
|
data Face =
|
|
|
|
Then users of our module would be able to use things of type Card, but wouldnt
be able to construct their own Cards and wouldnt be able to extract any of the suit/face
information stored in them.
If we wanted users of our module to be able to access all of this information, we
would have to specify it in the export list:
module Cards ( Card(Card),
Suit(Hearts,Spades,Diamonds,Clubs),
Face(Jack,Queen,King,Ace,Number),
...
)
6.2. IMPORTS
69
where
...
This can get frustrating if youre exporting datatypes with many constructors, so if
you want to export them all, you can simply write (..), as in:
module Cards ( Card(..),
Suit(..),
Face(..),
...
)
where
...
And this will automatically export all the constructors.
6.2 Imports
There are a few idiosyncracies in the module import system, but as long as you stay
away from the corner cases, you should be fine. Suppose, as before, you wrote a
module called Cards which you saved in the file Cards.hs. You are now writing
your poker module and you want to import all the definitions from the Cards module.
To do this, all you need to do is write:
module Poker
where
import Cards
This will enable to you use any of the functions, types and constructors exported
by the module Cards. You may refer to them simply by their name in the Cards
module (as, for instance, newDeck), or you may refer to them explicitely as imported
from Cards (as, for instance, Cards.newDeck). It may be the case that two module
export functions or types of the same name. In these cases, you can import one of
the modules qualified which means that you would no longer be able to simply use
the newDeck format but must use the longer Cards.newDeck format, to remove
ambiguity. If you wanted to import Cards in this qualified form, you would write:
import qualified Cards
Another way to avoid problems with overlapping function definitions is to import
only certain functions from modules. Suppose we knew the only function from Cards
that we wanted was newDeck, we could import only this function by writing:
CHAPTER 6. MODULES
70
71
6.4.1
Bird-scripts
In a Bird-style literate program, comments are default and code is introduced with a
leading greater-than sign (>). Everything else remains the same. For example, our
Hello World program would be written in Bird-style as:
This is a simple (literate!) Hello World program.
> module Main
>
where
All our main function does is print a string:
> main = putStrLn "Hello World"
Note that the spaces between the lines of code and the comments are necessary
(your compiler will probably complain if you are missing them). When compiled or
loaded in an interpreter, this program will have the exact same properties as the nonliterate version from Section 3.4.
CHAPTER 6. MODULES
72
6.4.2
LaTeX-scripts
LATEX is a text-markup language very popular in the academic community for publishing. If you are unfamiliar with LATEX, you may not find this section terribly useful.
Again, a literate Hello World program written in LATEX-style would look like:
This is another simple (literate!) Hello World program.
\begin{code}
module Main
where
\end{code}
All our main function does is print a string:
\begin{code}
main = putStrLn "Hello World"
\end{code}
In LATEX-style scripts, the blank lines are not necessary.
Chapter 7
Advanced Features
Discussion
74
Prelude> (+) 5 3
8
Prelude> (-) 5 3
2
Additionally, we can provide either of its argument to make a section. For example:
Prelude> (+5) 3
8
Prelude> (/3) 6
2.0
Prelude> (3/) 6
0.5
Non-infix functions can be made infix by enclosing them in backquotes (`). For
example:
Prelude> (+2) map [1..10]
[3,4,5,6,7,8,9,10,11,12]
75
76
f x =
let y = x+1
in y
where y = x+2
The value of f 5 is 6, not 7. Of course, I plead with you to never ever write
code that looks like this. No one should have to remember this rule and by shadowing
where-defined values in a let clause only makes your code difficult to understand.
In general, whether you should use let clauses or where clauses is largely a matter
of personal preference. Usually, the names you give to the subexpressions should be
sufficiently expressive that without reading their definitions any reader of your code
should be able to figure out what they do. In this case, where clauses are probably
more desirable because they allow the reader to see immediately what a function does.
However, in real life, values are often given cryptic names. In which case let clauses
may be better. Either is probably okay, though I think where clauses are more common.
eta reduction
Partial application is when you take a function which takes n arguments and you supply
it with < n of them. When discussing sections in Section 7.1, we saw a form of
partial application in which functions like + were partially applied. For instance, in
the expression map (+1) [1,2,3], the section (+1) is a partial application of +.
This is because + really takes two arguments, but weve only given it one.
Partial application is very common in function definitions and sometimes goes by
the name eta reduction. For instance, suppose we are writting a function lcaseString
which converts a whole string into lower case. We could write this as:
lcaseString s = map toLower s
Here, there is no partial application (though you could argue that applying no arguments to toLower could be considered partial application). However, we notice that
the application of s occurs at the end of both lcaseString and of map toLower.
In fact, we can remove it by performing eta reduction, to get:
lcaseString = map toLower
Now, we have a partial application of map: it expects a function and a list, but
weve only given it the function.
This all is related to type type of map, which is (a b) ([a] [b]), when
parentheses are all included. In our case, toLower is of type Char Char. Thus, if
we supply this function to map, we get a function of type [Char] [Char], as desired.
Now, consider the task of converting a string to lowercase and remove all non letter
characters. We might write this as:
77
point-free programming
$
function application
78
79
Exercises
Exercise 7.1 Convert the following functions into point-free style, if possible.
func1 x l = map (\y -> y*x) l
func2 f g l = filter f (map g l)
func3 f l = l ++ map f l
func4 l = map (\y -> y+2)
(filter (\z -> z elem [1..10])
(5:l))
func5 f l = foldr (\x y -> f (y,x)) 0 l
80
-- R G B components
We then want to write a function that will convert between something of type Color
and a triple of Ints, which correspond to the RGB values, respectively. Specifically, if
we see a Color which is Red, we want to return (255,0,0), since this is the RGB
value for red. So we write that (remember that piecewise function definitions are just
case statements):
colorToRGB Red = (255,0,0)
If we see a Color which is Orange, we want to return (255,128,0); and if we
see Yellow, we want to return (255,255,0), and so on. Finally, if we see a custom
color, which is comprised of three components, we want to make a triple out of these,
so we write:
colorToRGB
colorToRGB
colorToRGB
colorToRGB
colorToRGB
colorToRGB
colorToRGB
colorToRGB
Orange = (255,128,0)
Yellow = (255,255,0)
Green = (0,255,0)
Blue
= (0,0,255)
Purple = (255,0,255)
White = (255,255,255)
Black = (0,0,0)
(Custom r g b) = (r,g,b)
81
What is happening is this: we create a value, call it x, which has value Red. We
then apply this to colorToRGB. We check to see if we can match x against Red.
This match fails because according to the definition of Eq Color, Red is not equal
to Yellow. We continue down the definitions of colorToRGB and try to match
Yellow against Orange. This fails, too. We the try to match Yellow against
Yellow, which succeeds, so we use this function definition, which simply returns
the value (255,255,0), as expected.
Suppose instead, we used a custom color:
Color> colorToRGB (Custom 50 200 100)
(50,200,100)
We apply the same matching process, failing on all values from Red to Black.
We then get to try to match Custom 50 200 100 against Custom r g b. We
can see that the Custom part matches, so then we go see if the subelements match. In
the matching, the variables r, g and b are essentially wild cards, so there is no trouble
matching r with 50, g with 200 and b with 100. As a side-effect of this matching, r
gets the value 50, g gets the value 200 and b gets the value 100. So the entire match
succeeded and we look at the definition of this part of the function and bundle up the
triple using the matched values of r, g and b.
We can also write a function to check to see if a Color is a custom color or not:
isCustomColor (Custom _ _ _) = True
isCustomColor _ = False
When we apply a value to isCustomColor it tries to match that value against
. This match will succeed if the value is Custom x y z for any x,
Custom
y and z. The (underscore) character is a wildcard and will match anything, but will
not do the binding that would happen if you put a variable name there. If this match
succeeds, the function returns True; however, if this match fails, it goes on to the next
line, which will match anything and then return False.
For some reason we might want to define a function which tells us whether a given
color is bright or not, where my definition of bright is that one of its RGB components is equal to 255 (admittedly and arbitrary definition, but its simply an example).
We could define this function as:
isBright = isBright . colorToRGB
where isBright (255,_,_) = True
isBright (_,255,_) = True
isBright (_,_,255) = True
isBright _
= False
Lets dwell on this definition for a second. The isBright function is the composition of our previously defined function colorToRGB and a helper function isBright,
which tells us if a given RGB value is bright or not. We could replace the first line here
82
with isBright c = isBright (colorToRGB c) but there is no need to explicitly write the parameter here, so we dont. Again, this function composition style
of programming takes some getting used to, so I will try to use it frequently in this
tutorial.
The isBright helper function takes the RGB triple produced by colorToRGB.
It first tries to match it against (255, , ) which succeeds if the value has 255 in
its first position. If this match succeeds, isBright returns True and so does
isBright. The second and third line of definition check for 255 in the second and
third position in the triple, respectively. The fourth line, the fallthrough, matches everything else and reports it as not bright.
We might want to also write a function to convert between RGB triples and Colors.
We could simple stick everything in a Custom constructor, but this would defeat the
purpose; we want to use the Custom slot only for values which dont match the predefined colors. However, we dont want to allow the user to construct custom colors
like (600,-40,99) since these are invalid RGB values. We could throw an error if such
a value is given, but this can be difficult to deal with. Instead, we use the Maybe
datatype. This is defined (in the Prelude) as:
data Maybe a = Nothing
| Just a
The way we use this is as follows: our rgbToColor function returns a value of
type Maybe Color. If the RGB value passed to our function is invalid, we return
Nothing, which corresponds to a failure. If, on the other hand, the RGB value is
valid, we create the appropriate Color value and return Just that. The code to do this
is:
rgbToColor 255
0
0 = Just Red
rgbToColor 255 128
0 = Just Orange
rgbToColor 255 255
0 = Just Yellow
rgbToColor
0 255
0 = Just Green
rgbToColor
0
0 255 = Just Blue
rgbToColor 255
0 255 = Just Purple
rgbToColor 255 255 255 = Just White
rgbToColor
0
0
0 = Just Black
rgbToColor
r
g
b =
if 0 <= r && r <= 255 &&
0 <= g && g <= 255 &&
0 <= b && b <= 255
then Just (Custom r g b)
else Nothing
-- invalid RGB value
The first eight lines match the RGB arguments against the predefined values and,
if they match, rgbToColor returns Just the appropriate color. If none of these
matches, the last definition of rgbToColor matches the first argument against r, the
7.5. GUARDS
83
second against g and the third against b (which causes the side-effect of binding these
values). It then checks to see if these values are valid (each is greater than or equal to
zero and less than or equal to 255). If so, it returns Just (Custom r g b); if not,
it returns Nothing corresponding to an invalid color.
Using this, we can write a function that checks to see if a right RGB value is valid:
rgbIsValid r g b = rgbIsValid (rgbToColor r g b)
where rgbIsValid (Just _) = True
rgbIsValid _
= False
Here, we compose the helper function rgbIsValid with our function rgbToColor.
The helper function checks to see if the value returned by rgbToColor is Just anything (the wildcard). If so, it returns True. If not, it matches anything and returns
False.
Pattern matching isnt magic, though. You can only match against datatypes; you
cannot match against functions. For instance, the following is invalid:
f x = x + 1
g (f x) = x
Even though the intended meaning of g is clear (i.e., g x = x - 1), the compiler doesnt know in general that f has an inverse function, so it cant perform matches
like this.
7.5 Guards
Guards can be thought of as an extension to the pattern matching facility. They enable
you to allow piecewise function definitions to be taken according to arbitrary boolean
expressions. Guards appear after all arguments to a function but before the equals sign,
and are begun with a vertical bar. We could use guards to write a simple function which
returns a string telling you the result of comparing two elements:
comparison x y | x < y = "The first is less"
| x > y = "The second is less"
| otherwise = "They are equal"
You can read the vertical bar as such that. So we say that the value of comparison
x y such that x is less than y is The first is less. The value such that x is greater
than y is The second is less and the value otherwise is They are equal. The keyword otherwise is simply defined to be equal to True and thus matches anything
that falls through that far. So, we can see that this works:
84
Guards> comparison 5 10
"The first is less"
Guards> comparison 10 5
"The second is less"
Guards> comparison 7 7
"They are equal"
One thing to note about guards is that they are tested after pattern matching, not in
conjunction with pattern matching. This means that once a pattern matches, if none of
the guards succeed, further pattern matches will not be attempted. So, if we had instead
defined:
comparison2 x y | x < y = "The first is less"
| x > y = "The second is less"
comparison2 _ _ = "They are equal"
The intention would be that if both of the guards failed, it would fall through to
the final match and say that they were equal. This is not what happens, though.
Guards> comparison2 7 7
*** Exception: Guards.hs:8: Non-exhaustive patterns
in function comparison2
If we think about what is happening in the compiler this makes sense. When we
apply two sevens to comparison2, they are matched against x and y, which succeeds and the values are bound. Pattern matching then stops completely, since it has
succeeded. The guards are then activated and x and y are compared. Neither of the
guards succeeds, so an error is raised.
One nicety about guards is that where clauses are common to all guards. So another
possible definition for our isBright function from the previous section would be:
isBright2 c | r == 255 = True
| g == 255 = True
| b == 255 = True
| otherwise = False
where (r,g,b) = colorToRGB c
The function is equivalent to the previous version, but performs its calculation
slightly differently. It takes a color, c, and applies colorToRGB to it, yielding an
RGB triple which is matched (using pattern matching!) against (r,g,b). This match
succeeds and the values r, g and b are bound to their respective values. The first guard
checks to see if r is 255 and, if so, returns true. The second and third guard check g
and b against 255, respectively and return true if they match. The last guard fires as a
last resort and returns False.
85
7.6.1
The Eq Class
-- R G B components
86
7.6.2
The Show class is used to display arbitrary values as strings. This class has three
methods:
show :: Show a => a -> String
showsPrec :: Show a => Int -> a -> String -> String
showList :: Show a => [a] -> String -> String
A minimal complete definition is either show or showsPrec (we will talk about
showsPrec later its in there for efficiency reasons). We can define our Color
datatype to be an instance of Show with the following instance declaration:
instance
show
show
show
show
show
show
show
show
show
This declaration specifies exactly how to convert values of type Color to Strings.
Again, the first eight lines are identical and simply take a Color and produce a string.
87
The last line for handling custom colors matches out the RGB components and creates
a string by concattenating the result of showing the components individually (with
spaces in between and Custom at the beginning).
7.6.3
There are a few other important classes which I will mention briefly because either they
are commonly used or because we will be using them shortly. I wont provide example
instance declarations; how you can do this should be clear by now.
The Ord Class
The ordering class, the functions are:
compare :: Ord a => a -> a -> Ordering
(<=) :: Ord a => a -> a -> Bool
(>) :: Ord a => a -> a -> Bool
(>=) :: Ord a => a -> a -> Bool
(<) :: Ord a => a -> a -> Bool
min :: Ord a => a -> a -> a
max :: Ord a => a -> a -> a
The almost any of the functions alone is a minimal complete definition; it is recommended that you implement compare if you implement only one, though. This
function returns a value of type Ordering which is defined as:
data Ordering = LT | EQ | GT
So, for instance, we get:
Prelude> compare 5 7
LT
Prelude> compare 6 6
EQ
Prelude> compare 7 5
GT
In order to declare a type to be an instance of Ord you must already have declared
it an instance of Eq (in other words, Ord is a subclass of Eq more about this in
Section 8.4).
The Enum Class
The Enum class is for enumerated types; that is, for types where each element has a
successor and a predecessor. Its methods are:
88
a -> a
a -> a
a -> a
-> a
-> a
a
=> Integer -> a
All of these are obvious except for perhaps negate which is the unary minus.
That is, negate x means x.
The Read Class
The Read class is the opposite of the Show class. It is a way to take a string and read
in from it a value of arbitrary type. The methods for Read are:
readsPrec :: Read a => Int -> String -> [(a, String)]
readList :: String -> [([a], String)]
The minimal complete definition is readsPrec. The most important function
related to this is read, which uses readsPrec as:
read s = fst (head (readsPrec 0 s))
89
This will fail if parsing the string fails. You could define a maybeRead function
as:
maybeRead s =
case readsPrec 0 s of
[(a,_)] -> Just a
_ -> Nothing
How to write and use readsPrec directly will be discussed further in the examples.
7.6.4
Class Contexts
Suppose we are definition the Maybe datatype from scratch. The definition would be
something like:
data Maybe a = Nothing
| Just a
Now, when we go to write the instance declarations, for, say, Eq, we need to know
that a is an instance of Eq otherwise we cant write a declaration. We express this as:
instance Eq a => Eq (Maybe a) where
Nothing == Nothing = True
(Just x) == (Just x) == x == x
This first line can be read That a is an instance of Eq implies (=>) that Maybe a
is an instance of Eq.
7.6.5
Deriving Classes
Writing obvious Eq, Ord, Read and Show classes like these is tedious and should be
automated. Luckily for us, it is. If you write a datatype thats simple enough (almost
any datatype youll write unless you start writing fixed point types), the compiler can
automatically derive some of the most basic classes. To do this, you simply add a
deriving clause to after the datatype declaration, as in:
data Color
= Red
| ...
| Custom Int Int Int -- R G B components
deriving (Eq, Ord, Show, Read)
This will automatically create instances of the Color datatype of the named classes.
Similarly, the declaration:
90
7.7.1
Named Fields
---------
user name
local host
remote host
is guest?
is super user?
current directory
home directory
time connected
You could then write accessor functions, like (Ive only listed a few):
getUserName (Configuration un _ _ _ _ _ _ _) = un
getLocalHost (Configuration _ lh _ _ _ _ _ _) = lh
getRemoteHost (Configuration _ _ rh _ _ _ _ _) = rh
91
getIsGuest (Configuration _ _ _ ig _ _ _ _) = ig
...
You could also write update functions to update a single element. Of course, now
if you add an element to the configuration, or remove one, all of these functions now
have to take a different number of arguments. This is highly annoying and is an easy
place for bugs to slip in. However, theres a solution. We simply give names to the
fields in the datatype declaration, as follows:
data Configuration =
Configuration { username
localhost
remotehost
isguest
issuperuser
currentdir
homedir
timeconnected
}
::
::
::
::
::
::
::
::
String,
String,
String,
Bool,
Bool,
String,
String,
Integer
This will automatically generate the following accessor functions for us:
username :: Configuration -> String
localhost :: Configuration -> String
...
Moreover, it gives us very convenient update methods. Here is a short example
for a post working directory and change directory like functions that work on
Configurations:
changeDir :: Configuration -> String -> Configuration
changeDir cfg newDir =
-- make sure the directory exists
if directoryExists newDir
then -- change our current directory
cfg{currentdir = newDir}
else error "directory does not exist"
postWorkingDir :: Configuration -> String
-- retrieve our current directory
postWorkingDir cfg = currentdir cfg
So, in general, to update the field x in a datatype y to z, you write y{x=z}. You
can change more than one; each should be separated by commas, for instance, y{x=z,
a=b, c=d}.
92
You can of course continue to pattern match against Configurations as you did
before. The named fields are simply syntactic sugar; you can still write something like:
getUserName (Configuration un _ _ _ _ _ _ _) = un
But there is little reason to. Finally, you can pattern match against named fields as
in:
getHostData (Configuration {localhost=lh,remotehost=rh})
= (lh,rh)
This matches the variable lh against the localhost field on the Configuration
and the variable rh against the remotehost field on the Configuration. These
matches of course succeed. You could also constrain the matches by putting values
instead of variable names in these positions, as you would for standard datatypes.
You can create values of Configuration in the old way as shown in the first definition
below, or in the named-fields type, as shown in the second definition below:
initCFG =
Configuration "nobody" "nowhere" "nowhere"
False False "/" "/" 0
initCFG =
Configuration
{ username="nobody",
localhost="nowhere",
remotehost="nowhere",
isguest=False,
issuperuser=False,
currentdir="/",
homedir="/",
timeconnected=0 }
Though the second is probably much more understandable unless you litter your
code with comments.
7.8.1
Recall that the definition of the built-in Haskell list datatype is equivalent to:
data List a = Nil
| Cons a (List a)
93
With the exception that Nil is called [] and Cons x xs is called x:xs. This is
simply to make pattern matching easier and code smaller. Lets investigate how some
of the standard list functions may be written. Consider map. A definition is given
below:
map _ [] = []
map f (x:xs) = f x : map f xs
Here, the first line says that when you map across an empty list, no matter what the
function is, you get an empty list back. The second line says that when you map across
a list with x as the head and xs as the tail, the result is f applied to x consed onto the
result of mapping f on xs.
The filter can be defined similarly:
filter _ [] = []
filter p (x:xs) | p x = x : filter p xs
| otherwise = filter p xs
How this works should be clear. For an empty list, we return an empty list. For
a non empty list, we return the filter of the tail, perhaps with the head on the front,
depending on whether it satisfies the predicate p or not.
We can define foldr as:
foldr _ z [] = z
foldr f z (x:xs) = f x (foldr f z xs)
Here, the best interpretation is that we are replacing the empty list ([]) with a
particular value and the list constructor (:) with some function. On the first line, we
can see the replacement of [] for z. Using backquotes to make f infix, we can write
the second line as:
foldr f z (x:xs) = x f (foldr f z xs)
From this, we can directly see how : is being replaced by f.
Finally, foldl:
foldl _ z [] = z
foldl f z (x:xs) = foldl f (f z x) xs
This is slightly more complicated. Remember, z can be thought of as the current
state. So if were folding across a list which is empty, we simply return the current
state. On the other hand, if the list is not empty, its of the form x:xs. In this case, we
get a new state by appling f to the current state z and the current list element x and
then recursively call foldl on xs with this new state.
There is another class of functions: the zip and unzip functions, which respectively take multiple lists and make one or take one lists and split them apart. For
instance, zip does the following:
94
7.8.2
List Comprehensions
There is some syntactic sugar for dealing with lists whose elements are members of the
Enum class (see Section 7.6), such as Int or Char. If we want to create a list of all the
elements from 1 to 10, we can simply write:
Prelude> [1..10]
[1,2,3,4,5,6,7,8,9,10]
We can also introduce an amount to step by:
Prelude> [1,3..10]
[1,3,5,7,9]
Prelude> [1,4..10]
[1,4,7,10]
These expressions are short hand for enumFromTo and enumFromThenTo, respectively. Of course, you dont need to specify an upper bound. Try the following
(but be ready to hit Control+C to stop the computation!):
95
Prelude> [1..]
[1,2,3,4,5,6,7,8,9,10,11,12{Interrupted!}
Probably yours printed a few thousand more elements than this. As we said before,
Haskell is lazy. That means that a list of all numbers from 1 on is perfectly well formed
and thats exactly what this list is. Of course, if you attempt to print the list (which
were implicitly doing by typing it in the interpreter), it wont halt. But if we only
evaluate an initial segment of this list, were fine:
Prelude> take 3 [1..]
[1,2,3]
Prelude> take 3 (drop 5 [1..])
[6,7,8]
This comes in useful if, say, we want to assign an ID to each element in a list.
Without laziness wed have to write something like this:
assignID :: [a] -> [(a,Int)]
assignID l = zip l [1..length l]
Which means that the list will be traversed twice. However, because of laziness,
we can simply write:
assignID l = zip l [1..]
And well get exactly what we want. We can see that this works:
Prelude> assignID "hello"
[(h,1),(e,2),(l,3),(l,4),(o,5)]
Finally, there is some useful syntactic sugar for map and filter, based on standard set-notation in mathematics. In math, we would write something like {f (x)|x
s p(x)} to mean the set of all values of f when applied to elements of s which satisfy
p. This is equivalent to the Haskell statement map f (filter p s). However, we
can also use more math-like notation and write [f x | x <- s, p x]. While in
math the ordering of the statements on the side after the pipe is free, it is not so in
Haskell. We could not have put p x before x <- s otherwise the compiler wouldnt
know yet what x was. We can use this to do simple string processing. Suppose we
want to take a string, remove all the lower-case letters and convert the rest of the letters
to upper case. We could do this in either of the following two equivalent ways:
Prelude> map toLower (filter isUpper "Hello World")
"hw"
Prelude> [toLower x | x <- "Hello World", isUpper x]
"hw"
96
These two are equivalent, and, depending on the exact functions youre using, one
might be more readable than the other. Theres more you can do here, though. Suppose
you want to create a list of pairs, one for each point between (0,0) and (5,7) below the
diagonal. Doing this manually with lists and maps would be cumbersome and possibly
difficult to read. It couldnt be easier than with list comprehensions:
Prelude> [(x,y) | x <- [1..5], y <- [x..7]]
[(1,1),(1,2),(1,3),(1,4),(1,5),(1,6),(1,7),(2,2),(2,3),
(2,4),(2,5),(2,6),(2,7),(3,3),(3,4),(3,5),(3,6),(3,7),
(4,4),(4,5),(4,6),(4,7),(5,5),(5,6),(5,7)]
If you reverse the order of the x <- and y <- clauses, the order in which the
space is traversed will be reversed (of course, in that case, y could no longer depend
on x and you would need to make x depend on y but this is trivial).
7.9 Arrays
Lists are nice for many things. It is easy to add elements to the beginning of them and
to manipulate them in various ways that change the length of the list. However, they are
bad for random access, having average complexity O(n) to access an arbitrary element
(if you dont know what O(. . . ) means, you can either ignore it or take a quick detour
and read Appendix A, a two-page introduction to complexity theory). So, if youre
willing to give up fast insertion and deletion because you need random access, you
should use arrays instead of lists.
In order to use arrays you must import the Array module. There are a few
methods for creating arrays, the array function, the listArray function, and the
accumArray function. The array function takes a pair which is the bounds of
the array, and an association list which specifies the initial values of the array. The
listArray function takes bounds and then simply a list of values. Finally, the
accumArray function takes an accumulation function, an initial value and an association list and accumulates pairs from the list into the array. Here are some examples of
arrays being created:
Arrays> array (1,5) [(i,2*i) | i <- [1..5]]
array (1,5) [(1,2),(2,4),(3,6),(4,8),(5,10)]
Arrays> listArray (1,5) [3,7,5,1,10]
array (1,5) [(1,3),(2,7),(3,5),(4,1),(5,10)]
Arrays> accumArray (+) 2 (1,5) [(i,i) | i <- [1..5]]
array (1,5) [(1,3),(2,4),(3,5),(4,6),(5,7)]
When arrays are printed out (via the show function), they are printed with an association list. For instance, in the first example, the association list says that the value of
the array at 1 is 2, the value of the array at 2 is 4, and so on.
You can extract an element of an array using the ! function, which takes an array
and an index, as in:
97
98
In all these cases, the type key must be an instance of Ord (and hence also an
instance of Eq).
There are also function listToFM and fmToList to convert lists to and from
finite maps. Try the following:
Prelude> :m FiniteMap
FiniteMap> let fm = listToFM
[(a,5),(b,10),(c,1),(d,2)]
FiniteMap> let myFM = addToFM fm e 6
FiniteMap> fmToList fm
[(a,5),(b,10),(c,1),(d,2)]
FiniteMap> fmToList myFM
[(a,5),(b,10),(c,1),(d,2),(e,6)]
FiniteMap> lookupFM myFM e
Just 6
FiniteMap> lookupFM fm e
Nothing
You can also experiment with the other commands. Note that you cannot show a
finite map, as they are not instances of Show:
FiniteMap> show myFM
7.11. LAYOUT
99
<interactive>:1:
No instance for (Show (FiniteMap Char Integer))
arising from use of show at <interactive>:1
In the definition of it: show myFM
In order to inspect the elements, you first need to use fmToList.
7.11 Layout
7.12 The Final Word on Lists
You are likely tired of hearing about lists at this point, but they are so fundamental to
Haskell (and really all of functional programming) that it would be terrible not to talk
about them some more.
It turns out that foldr is actually quite a powerful function: it can compute an
primitive recursive function. A primitive recursive function is essentially one which
can be calculated using only for loops, but not while loops.
In fact, we can fairly easily define map in terms of foldr:
map2 f = foldr (\a b -> f a : b) []
Here, b is the accumulator (i.e., the result list) and a is the element being currently
considered. In fact, we can simplify this definition through a sequence of steps:
==>
==>
==>
==>
foldr
foldr
foldr
foldr
foldr
(\a b -> f a : b) []
(\a b -> (:) (f a) b) []
(\a -> (:) (f a)) []
(\a -> ((:) . f) a) []
((:) . f) []
This is directly related to the fact that foldr (:) [] is the identity function on
lists. This is because, as mentioned before, foldr f z can be thought of as replacing
the [] in lists by z and the : by f. In this case, were keeping both the same, so it is
the identity function.
In fact, you can convert any function of the following style into a foldr:
myfunc [] = z
myfunc (x:xs) = f x (myfunc xs)
By writing the last line with f in infix form, this should be obvious:
myfunc [] = z
myfunc (x:xs) = x f (myfunc xs)
primitive recursive
100
Clearly, we are just replacing [] with z and : with f. Consider the filter
function:
filter p
filter p
if p x
then
else
[] = []
(x:xs) =
x : filter p xs
filter p xs
This function also follows the form above. Based on the first line, we can figure
out that z is supposed to be [], just like in the map case. Now, suppose that we call
the result of calling filter p xs simply b, then we can rewrite this as:
filter p [] = []
filter p (x:xs) =
if p x then x : b else b
Given this, we can transform filter into a fold:
filter p = foldr (\a b -> if p a then a:b else b) []
Lets consider a slightly more complicated function: ++. The definition for ++ is:
(++) []
ys = ys
(++) (x:xs) ys = x : (xs ++ ys)
Now, the question is whether we can write this in fold notation. First, we can apply
eta reduction to the first line to give:
(++) [] = id
Through a sequence of steps, we can also eta-reduce the second line:
==>
==>
==>
(++)
(++)
(++)
(++)
(x:xs)
(x:xs)
(x:xs)
(x:xs)
ys = x : ((++) xs ys)
ys = (x:) ((++) xs ys)
ys = ((x:) . (++) xs) ys
= (x:) . (++) xs
101
(++)
(++)
(++)
(++)
(++)
(++)
=
=
=
=
=
=
foldr
foldr
foldr
foldr
foldr
foldr
This final version is point free, though not necessarily understandable. Presumbably the original version is clearer.
As a final example, consider concat. We can write this as:
concat []
= []
concat (x:xs) = x ++ concat xs
It should be immediately clear that the z element for the fold is [] and that the
recursive function is ++, yielding:
concat = foldr (++) []
Exercises
Exercise 7.2 The function and takes a list of booleans and returns True if and only
if all of them are True. It also returns True on the empty list. Write this function in
terms of foldr.
Exercise 7.3 The function concatMap behaves such that concatMap f is the same
as concat . map f. Write this function in terms of foldr.
102
Chapter 8
Advanced Types
As youve probably ascertained by this point, the type system is integral to Haskell.
While this chapter is called Advanced Types, you will probably find it to be more
general than that and it must not be skipped simply because youre not interested in the
type system.
Type synonyms exist in Haskell simply for convenience: their removal would not make
Haskell any less powerful.
Consider the case when you are constantly dealing with lists of three-dimensional
points. For instance, you might have a function with type [(Double, Double, Double)] Double [(Double, Double, Double)].
Since you are a good software engineer, you want to place type signatures on all your
top-level functions. However, typing [(Double, Double, Double)] all the time gets very
tedious. To get around this, you can define a type synonym:
type List3D = [(Double,Double,Double)]
Now, the type signature for your functions may be written List3D Double List3D.
We should note that type synonyms cannot be self-referential. That is, you cannot
have:
type BadType = Int -> BadType
This is because this is an infinite type. Since Haskell removes type synonyms
very early on, any instance of BadType will be replaced by Int BadType, which
will result in an infinite loop.
Type synonyms can also be parameterized. For instance, you might want to be able
to change the types of the points in the list of 3D points. For this, you could define:
type List3D a = [(a,a,a)]
103
104
Then your references to [(Double, Double, Double)] would become List3D Double.
8.2 Newtypes
Consider the problem in which you need to have a type which is very much like Int, but
its ordering is defined differently. Perhaps you wish to order Ints first by even numbers
then by odd numbers (that is, all odd numbers are greater than any even number and
within the odd/even subsets, ordering is standard).
Unfortunately, you cannot define a new instance of Ord for Int because then
Haskell wont know which one to use. What you want is to define a type which is
isomorphic to Int.
NOTE Isomorphic is a common term in mathematics which basically means structurally identical. For instance, in graph theory, if you
have two graphs which are identical except they have different labels on
the nodes, they are isomorphic. In our context, two types are isomorphic
if they have the same underlying structure.
One way to do this would be to define a new datatype:
data MyInt = MyInt Int
We could then write appropriate code for this datatype. The problem (and this is
very subtle) is that this type is not truly isomorphic to Int: it has one more value. When
we think of the type Int, we usually think that it takes all values of integers, but it really
has one more value: | (pronounced bottom), which is used to represent erroneous or
undefined computations. Thus, MyInt has not only values MyInt 0, MyInt 1 and
so on, but also MyInt | . However, since datatypes can themselves be undefined, it
has an additional value: | which differs from MyInt | and this makes the types
non-isomorphic. (See Section ?? for more information on bottom.)
Disregarding that subtlety, there may be efficiency issues with this representation:
now, instead of simply storing an integer, we have to store a pointer to an integer and
have to follow that pointer whenever we need the value of a MyInt.
To get around these problems, Haskell has a newtype construction. A newtype is a
cross between a datatype and a type synonym: it has a constructor like a datatype, but
it can have only one constructor and this constructor can have only one argument. For
instance, we can define:
newtype MyInt = MyInt Int
But we cannot define any of:
newtype Bad1 = Bad1a Int | Bad1b Double
newtype Bad2 = Bad2 Int Double
8.3. DATATYPES
105
Of course, the fact that we cannot define Bad2 as above is not a big issue: we can
simply define the following by pairing the types:
newtype Good2 = Good2 (Int,Double)
Now, suppose weve defined MyInt as a newtype. This enables use to write our
desired instance of Ord as:
instance Ord MyInt where
MyInt i < MyInt j
| odd i && odd j = i < j
| even i && even j = i < j
| even i
= True
| otherwise
= False
where odd x = (x mod 2) == 0
even = not . odd
Like datatype, we can still derive classes like Show and Eq over newtypes (in fact,
Im implicitly assuming we have derived Eq over MyInt where is my assumption in
the above code?).
Moreover, in recent versions of GHC (see Section 2.2), on newtypes, you are allowed to derive any class of which the base type (in this case, Int) is an instance. For
example, we could derive Num on MyInt to provide arithmetic functions over it.
Pattern matching over newtypes is exactly as in datatypes. We can write constructor
and destructor functions for MyInt as follows:
mkMyInt i = MyInt i
unMyInt (MyInt i) = i
8.3 Datatypes
Weve already seen datatypes used in a variety of contexts. This section concludes
some of the discussion and introduces some of the common datatypes in Haskell. It
also provides a more theoretical underpinning to what datatypes actually are.
8.3.1
Strict Fields
One of the great things about Haskell is that computation is performed lazily. However,
sometimes this leads to inefficiencies. One way around this problem is to use datatypes
with strict fields. Before we talk about the solution, lets spend some time to get a
bit more comfortable with how bottom works in to the picture (for more theory, see
Section ??).
Suppose weve defined the unit datatype (this one of the simplest datatypes you can
define):
106
deriving Show
8.3. DATATYPES
main = do
[cmd] <case cmd
"a" ->
"b" ->
"c" ->
"d" ->
"e"
"f"
"g"
"h"
->
->
->
->
getArgs
of
printJust
printJust
printJust
printJust
107
undefined
Nothing
(Just undefined)
(Just ())
printSJust undefined
printSJust SNothing
printSJust (SJust undefined)
printSJust (SJust ())
108
\% ./strict h
Just ()
The thing worth noting here is the difference between cases c and g. In the
c case, the Just is printed, because this is printed before the undefined value is
evaluated. However, in the g case, since the constructor is strict, as soon as you
match the SJust, you also match the value. In this case, the value is undefined, so the
whole thing fails before it gets a chance to do anything.
8.4 Classes
We have already encountered type classes a few times, but only in the context of previously existing type classes. This section is about how to define your own. We will
begin the discussion by talking about Pong and then move on to a useful generalization
of computations.
8.4.1
Pong
The discussion here will be motivated by the construction of the game Pong (see Appendix ?? for the full code). In Pong, there are three things drawn on the screen: the
two paddles and the ball. While the paddles and the ball are different in a few respects,
they share many commonalities, such as position, velocity, acceleration, color, shape,
and so on. We can express these commonalities by defining a class for Pong entities,
which we call Entity. We make such a definition as follows:
class Entity a where
getPosition :: a -> (Int,Int)
getVelocity :: a -> (Int,Int)
getAcceleration :: a -> (Int,Int)
getColor :: a -> Color
getShape :: a -> Shape
This code defines a typeclass Entity. This class has five methods: getPosition,
getVelocity, getAcceleration, getColor and getShape with the corresponding types.
The first line here uses the keyword class to introduce a new typeclass. We can
read this typeclass definition as There is a typeclass Entity; a type a is an instance
of Entity if it provides the following five functions: . . . . To see how we can write an
instance of this class, let us define a player (paddle) datatype:
data Paddle =
Paddle { paddlePosX,
paddleVelX,
paddleAccX,
paddleColor
paddlePosY,
paddleVelY,
paddleAccY :: Int,
:: Color,
8.4. CLASSES
109
paddleHeight :: Int,
playerNumber :: Int }
8.4.2
Computations
Lets think back to our original motivation for defining the Maybe datatype from Section ??. We wanted to be able to express that functions (i.e., computations) can fail.
Let us consider the case of performing search on a graph. Allow us to take a small
aside to set up a small graph library:
data Graph v e = Graph [(Int,v)] [(Int,Int,e)]
The Graph datatype takes two type arguments which correspond to vertex and edge
labels. The first argument to the Graph constructor is a list (set) of vertices; the second
is the list (set) of edges. We will assume these lists are always sorted and that each
vertex has a unique id and that there is at most one edge between any two vertices.
Suppose we want to search for a path between two vertices. Perhaps there is no
path between those vertices. To represent this, we will use the Maybe datatype. If
it succeeds, it will return the list of vertices traversed. Our search function could be
written (naively) as follows:
search :: Graph v e -> Int -> Int -> Maybe [Int]
search g@(Graph vl el) src dst
| src == dst = Just [src]
110
This algorithm works as follows (try to read along): to search in a graph g from
src to dst, first we check to see if these are equal. If they are, we have found our
way and just return the trivial solution. Otherwise, we want to traverse the edge-list.
If were traversing the edge-list and it is empty, weve failed, so we return Nothing.
Otherwise, were looking at an edge from u to v. If u is our source, then we consider
this step and recursively search the graph from v to dst. If this fails, we try the rest of
the edges; if this succeeds, we put our current position before the path found and return.
If u is not our source, this edge is useless and we continue traversing the edge-list.
This algorithm is terrible: namely, if the graph contains cycles, it can loop indefinitely. Nevertheless, it is sufficent for now. Be sure you understand it well: things only
get more complicated.
Now, there are cases where the Maybe datatype is not sufficient: perhaps we wish
to include an error message together with the failure. We could define a datatype to
express this as:
data Failable a = Success a | Fail String
Now, failures come with a failure string to express what went wrong. We can
rewrite our search function to use this datatype:
search2 :: Graph v e -> Int -> Int -> Failable [Int]
search2 g@(Graph vl el) src dst
| src == dst = Success [src]
| otherwise = search el
where search [] = Fail "No path"
search ((u,v,_):es)
| src == u =
case search2 g v dst of
Success p -> Success (u:p)
_
-> search es
| otherwise = search es
This code is a straightforward translation of the above.
There is another option for this computation: perhaps we want not just one path,
but all possible paths. We can express this as a function which returns a list of lists of
vertices. The basic idea is the same:
8.4. CLASSES
111
112
augment
augment
combine
combine
(Just x) f = f x
Nothing _ = Nothing
Nothing y = y
x _ = x
Here, success is represented with Just and failure ignores its argument and
returns Nothing. The combine function takes the first success we found and ignores
the rest. The function augment checks to see if we succeeded before (and thus had
a Just something) and, if we did, applies f to it. If we failed before (and thus had a
Nothing), we ignore the function and return Nothing.
instance Computation Failable where
success = Success
failure = Fail
augment (Success x) f = f x
augment (Fail s) _ = Fail s
combine (Fail _) y = y
combine x _ = x
These definitions are obvious. Finally:
instance Computation [] where
success a = [a]
failure = const []
augment l f = concat (map f l)
combine = (++)
Here, the value of a successful computation is a singleton list containing that value.
Failure is represented with the empty list and to combine previous successes we simply
catentate them. Finally, augmenting a computation amounts to mapping the function
across the list of previous computations and concatentate them. we apply the function
to each element in the list and then concatenate the results.
Using these computations, we can express all of the above versions of search as:
searchAll g@(Graph vl el) src dst
| src == dst = success [src]
| otherwise = search el
where search [] = failure "no path"
search ((u,v,_):es)
| src == u = (searchAll g v dst augment
(success . (u:)))
combine search es
| otherwise = search es
8.5. INSTANCES
113
In this, we see the uses of all the functions from the class Computation.
If youve understood this discussion of computations, you are in a very good position as you have understood the concept of monads, probably the most difficult concept
in Haskell. In fact, the Computation class is almost exactly the Monad class, except that success is called return, failure is called fail and augment is
called >>= (read bind). The combine function isnt actually required by monads,
but is found in the MonadPlus class for reasons which will become obvious later.
If you didnt understand everything here, read through it again and then wait for
the proper discussion of monads in Chapter 9.
8.5 Instances
We have already seen how to declare instances of some simple classes; allow us to
consider some more advanced classes here. There is a Functor class defined in the
Functor module.
NOTE The name functor, like monad comes from category theory. There, a functor is like a function, but instead of mapping elements
to elements, it maps structures to structures.
The definition of the functor class is:
class Functor f where
fmap :: (a -> b) -> f a -> f b
The type definition for fmap (not to mention its name) is very similar to the function map over lists. In fact, fmap is essentially a generalization of map to arbitrary
structures (and, of course, lists are already instances of Functor). However, we can
also define other structures to be instances of functors. Consider the following datatype
for binary trees:
data BinTree a = Leaf a
| Branch (BinTree a) (BinTree a)
We can immediately identify that the BinTree type essentially raises a type a into
trees of that type. There is a naturally associated functor which goes along with this
raising. We can write the instance:
instance Functor BinTree where
fmap f (Leaf a) = Leaf (f a)
fmap f (Branch left right) =
Branch (fmap f left) (fmap f right)
114
Now, weve seen how to make something like BinTree an instance of Eq by using
the deriving keyword, but here we will do it by hand. We want to make BinTree as
instances of Eq but obviously we cannot do this unless a is itself an instance of Eq.
We can specify this dependence in the instance declaration:
instance Eq a => Eq (BinTree a) where
Leaf a == Leaf b = a == b
Branch l r == Branch l r = l == l && r == r
_ == _ = False
The first line of this can be read if a is an instance of Eq, then BinTree a is also
an instance of Eq. We then provide the definitions. If we did not include the Eq a
=> part, the compiler would complain because were trying to use the == function on
as in the second line.
The Eq a => part of the definition is called the context. We should note that
there are some restrictions on what can appear in the context and what can appear in
the declaration. For instance, were not allowed to have instance declarations that dont
contain type constructors on the right hand side. To see why, consider the following
declarations:
class MyEq a where
myeq :: a -> a -> Bool
instance Eq a => MyEq a where
myeq = (==)
As it stands, there doesnt seem to be anything wrong with this definition. However,
if elsewhere in a program we had the definition:
instance MyEq a => Eq a where
(==) = myeq
In this case, if were trying to establish if some type is an instance of Eq, we could
reduce it to trying to find out if that type is an instance of MyEq, which we could
in turn reduce to trying to find out if that type is an instance of Eq, and so on. The
compiler protects itself against this by refusing the first instance declaration.
This is commonly known as the closed-world assumption. That is, were assuming,
when we write a definition like the first one, that there wont be any declarations like
the second. However, this assumption is invalid because theres nothing to prevent the
second declaration (or some equally evil declaration). The closed world assumption
can also bite you in cases like:
class OnlyInts a where
foo :: a -> a -> Bool
8.6. KINDS
115
8.6 Kinds
Let us take a moment and think about what types are available in Haskell. We have
simple types, like Int, Char, Double and so on. We then have type constructors like
Maybe which take a type (like Char) and produce a new type, Maybe Char. Similarly,
the type constructor [] (lists) takes a type (like Int) and produces [Int]. We have more
complex things like (function arrow) which takes two types (say Int and Bool) and
produces a new type Int Bool.
In a sense, these types themselves have type. Types like Int have some sort of basic
type. Types like Maybe have a type which takes something of basic type and returns
something of basic type. And so forth.
Talking about the types of types becomes unwieldy and highly ambiguous, so we
call the types of types kinds. What we have been calling basic types have kind
*. Something of kind * is something which can have an actual value. There is also a
single kind constructor, with which we can build more complex kinds.
Consider Maybe. This takes something of kind * and produces something of kind
*. Thus, the kind of Maybe is * -> *. Recall the definition of Pair from Section 4.5.1:
data Pair a b = Pair a b
Here, Pair is a type constructor which takes two arguments, each of kind * and
produces a type of kind *. Thus, the kind of Pair is * -> (* -> *). However, we
again assume associativity so we just write * -> * -> *.
Let us make a slightly strange datatype definition:
data Strange c a b =
MkStrange (c a) (c b)
Before we analyze the kind of Strange, lets think about what it does. It is essentially a pairing constructor, though it doesnt pair actual elements, but elements within
another constructor. For instance, think of c as Maybe. Then MkStrange pairs Maybes
of the two types a and b. However, c need not be Maybe but could instead by [], or
many other things.
116
What do we know about c, though? We know that it must have kind * -> *. This
is because we have c a on the right hand side. The type variables a and b each have
kind * as before. Thus, the kind of Strange is (* -> *) -> * -> * -> *. That
is, it takes a constructor (c) of kind * -> * together with two types of kind * and
produces something of kind *.
A question may arise regarding how we know a has kind * and not some other
kind k. In fact, the inferred kind for Strange is (k -> *) -> k -> k -> *.
However, this requires polymorphism on the kind level, which is too complex, so we
make a default assumption that k = *.
NOTE There are extensions to GHC which allow you to specify the
kind of constructors directly. For instance, if you wanted a different kind,
you could write this explicitly:
data Strange (c :: (* -> *) -> *) a b = MkStrange (c a) (c b)
to give a different kind to Strange.
The notation of kinds suggests that we can perform partial application, as we can
for functions. And, in fact, we can. For instance, we could have:
type MaybePair = Strange Maybe
The kind of MaybePair is, not surprisingly, * -> * -> *.
We should note here that all of the following definitions are acceptable:
type MaybePair1
= Strange Maybe
type MaybePair2 a
= Strange Maybe a
type MaybePair3 a b = Strange Maybe a b
These all appear to be the same, but they are in fact not identical as far as Haskells
type system is concerned. The following are all valid type definitions using the above:
type MaybePair1a = MaybePair1
type MaybePair1b = MaybePair1 Int
type MaybePair1c = MaybePair1 Int Double
type MaybePair2b = MaybePair2 Int
type MaybePair2c = MaybePair2 Int Double
type MaybePair3c = MaybePair3 Int Double
But the following are not valid:
117
118
Chapter 9
Monads
The most difficult concept to master, while learning Haskell, is that of understanding
and using monads. We can distinguish two subcomponents here: (1) learning how
to use existing monads and (2) learning how to write new ones. If you want to use
Haskell, you must learn to use existing monads. On the other hand, you will only need
to learn to write your own monads if you want to become a super Haskell guru. Still,
if you can grasp writing your own monads, programming in Haskell will be much more
pleasant.
So far weve seen two uses of monads. The first use was IO actions: Weve seen
that, by using monads, we can abstract get away from the problems plaguing the RealWorld solution to IO presented in Chapter 5. The second use was representing different
types of computations in Section 8.4.2. In both cases, we needed a way to sequence
operations and saw that a sufficient definition (at least for computations) was:
class Computation c where
success :: a -> c a
failure :: String -> c a
augment :: c a -> (a -> c b) -> c b
combine :: c a -> c a -> c a
Lets see if this definition will enable us to also perform IO. Essentially, we need
a way to represent taking a value out of an action and performing some new operation
on it (as in the example from Section 4.4.3, rephrased slightly):
main = do
s <- readFile "somefile"
putStrLn (show (f s))
But this is exactly what augment does. Using augment, we can write the above
code as:
119
computations
CHAPTER 9. MONADS
120
bind
then
m where
:: a -> m
:: String
:: m a ->
:: m a ->
a
-> m a
(a -> m b) -> m b
m b -> m b
9.1 Do Notation
syntactic sugar
We have hinted that there is a connection between monads and the do notation. Here,
we make that relationship concrete. There is actually nothing magic about the do
notation it is simply syntactic sugar for monadic operations.
As we mentioned earlier, using our Computation class, we could define our
above program as:
main =
readFile "somefile" augment \s ->
putStrLn (show (f s))
But we now know that augment is called >>= in the monadic world. Thus, this
program really reads:
main =
readFile "somefile" >>= \s ->
putStrLn (show (f s))
9.1. DO NOTATION
121
And this is completely valid Haskell at this point: if you defined a function f ::
Show a => String -> a, you could compile and run this program)
This suggests that we can translate:
x <- f
g x
into f >>= \x -> g x. This is exactly what the compiler does. Talking about
do becomes easier if we do not use implicit layout (see Section ?? for how to do this).
There are four translation rules:
1. do {e} e
2. do {e; es} e >> do {es}
3. do {let decls; es} let decls in do {es}
4. do {p <- e; es} let ok p = do {es} ; ok
in e >>= ok
= fail "..."
Translation Rule 1
The first translation rule, do {e} e, states (as we have stated before) that when
performing a single action, having a do or not is irrelevant. This is essentially the base
case for an inductive definition of do. The base case has one action (namely e here);
the other three translation rules handle the cases where there is more than one action.
Translation Rule 2
This states that do {e; es} e >> do {es}. This tells us what to do if we have
an action (e) followed by a list of actions (es). Here, we make use of the >> function,
defined earlier. This rule simple states that to do {e; es}, we first perform the action
e, throw away the result, and then do es.
For instance, if e is putStrLn s for some string s, then the translation of do
{e; es} is to perform e (i.e., print the string) and then do es. This is clearly what
we want.
Translation Rule 3
This states that do {let decls; es} let decls in do {es}. This rule
tells us how to deal with lets inside of a do statement. We lift the declarations within
the let out and do whatever comes after the declarations.
let
CHAPTER 9. MONADS
122
Translation Rule 4
This states that do {p <- e; es} let ok p = do {es} ; ok
= fail
"..." in e >>= ok. Again, it is not exactly obvious what is going on here. However, an alternate formulation of this rule, which is roughly equivalent, is: do {p <e; es} e >>= \p -> es. Here, it is clear what is happening. We run the
action e, and then send the results into es, but first give the result the name p.
The reason for the complex definition is that p doesnt need to simply be a variable;
it could be some complex pattern. For instance, the following is valid code:
foo = do (a:b:c:x:xs) <- getLine
putStrLn (x:xs)
In this, were assuming that the results of the action getLine will begin with
the string abc and will have at least one more character. The question becomes
what should happen if this pattern match fails. The compiler could simply throw an
error, like usual, for failed pattern matches. However, since were within a monad, we
have access to a special fail function, and wed prefer to fail using that function,
rather than the catch all error function. Thus, the translation, as defined, allows
the compiler to fill in the ... with an appropriate error message about the pattern
matching having failed. Apart from this, the two definitions are equivalent.
9.2 Definition
monad laws
There are three rules that all monads must obey called the Monad Laws (and it is up
to you to ensure that your monads obey these rules) :
1. return a >>= f f a
2. f >>= return f
3. f >>= (\x -> g x >>= h) (f >>= g) >>= h
Lets look at each of these individually:
Law 1
This states that return a >>= f f a. Suppose we think about monads as computations. This means that if we create a trivial computation that simply returns the
value a regardless of anything else (this is the return a part); and then bind it together with some other computation f, then this is equivalent to simply performing the
computation f on a directly.
For example, suppose f is the function putStrLn and a is the string Hello
World. This rule states that binding a computation whose result is Hello World
to putStrLn is the same as simply printing it to the screen. This seems to make
sense.
In do notation, this law states that the following two programs are equivalent:
9.2. DEFINITION
123
law1a = do
x <- return a
f x
law1b = do
f a
Law 2
The second monad law states that f >>= return f. That is, f is some computation, the law states that if we perform the computation f and then pass the result on to
the trivial return function, then all we have done is to perform the computation.
That this law must hold should be obvious. To see this, think of f as getLine
(reads a string from the keyboard). This law states that reading a string and then returning the value read is exactly the same as just reading the string.
In do notation, the law states that the following two programs are equivalent:
law2a = do
x <- f
return x
law2b = do
f
Law 3
This states that f >>= (\x -> g x >>= h) (f >>= g) >>= h. At first
glance, this law is not as easy to grasp as the other two. It is essentially an associativity
law for monads.
NOTE Outside the world of monads, a function is associative if
(f g) h = f (g h). For instance, + and * are associative, since
bracketing on these functions doesnt make a difference. On the other
hand, - and / are not associative since, for example, 5 (3 1) 6=
(5 3) 1.
If we throw away the messiness with the lambdas, we see that this law states: f
>>= (g >>= h) (f >>= g) >>= h. The intuition behind this law is that
when we string together actions, it doesnt matter how we group them.
For a concrete example, take f to be getLine. Take g to be an action which takes
a value as input, prints it to the screen, reads another string via getLine, and then
returns that newly read string. Take h to be putStrLn.
Lets consider what (\x -> g x >>= h) does. It takes a value called x, and
runs g on it, feeding the results into h. In this instance, this means that its going to
associative
124
CHAPTER 9. MONADS
take a value, print it, read another value and then print that. Thus, the entire left hand
side of the law first reads a string and then does what weve just described.
On the other hand, consider (f >>= g). This action reads a string from the
keyboard, prints it, and then reads another string, returning that newly read string as a
result. When we bind this with h as on the right hand side of the law, we get an action
that does the action described by (f >>= g), and then prints the results.
Clearly, these two actions are the same.
While this explanation is quite complicated, and the text of the law is also quite
complicated, the actual meaning is simple: if we have three actions, and we compose
them in the same order, it doesnt matter where we put the parentheses. The rest is just
notation.
In do notation, says that the following two programs are equivalent:
law3a = do
x <- f
do g <- x
h y
law3b = do
y <- do x <- f
g x
h y
125
CHAPTER 9. MONADS
126
a with state st. The second is a function that takes this a and produces something of
type b also with the same state. The result of bindState is essentially the result of
transforming the a into a b.
The definition of bindState takes an initial state, st. It first applies this to the
State st a argument called m. This gives back a new state st and a value a. It
then lets the function k act on a, producing something of type State st b, called
m. We finally run m with the new state st.
We write a new function, mapTreeStateM and give it the type:
mapTreeStateM :: (a -> State st b) -> Tree a -> State st (Tree b)
Using these plumbing functions (returnState and bindState) we can
write this function without ever having to explicitly talk about the state:
mapTreeStateM f (Leaf a) =
f a bindState \b ->
returnState (Leaf b)
mapTreeStateM f (Branch lhs rhs) =
mapTreeStateM f lhs bindState \lhs ->
mapTreeStateM f rhs bindState \rhs ->
returnState (Branch lhs rhs)
In the Leaf case, we apply f to a and then bind the result to a function that takes
the result and returns a Leaf with the new value.
In the Branch case, we recurse on the left-hand-side, binding the result to a function that recurses on the right-hand-side, binding that to a simple function that returns
the newly created Branch.
As you have probably guessed by this point, State st is a monad, returnState
is analogous to the overloaded return method, and bindState is analogous to the
overloaded >>= method. In fact, we can verify that State st a obeys the monad
laws:
Law 1 states: return a >>= f f a. Lets calculate on the left hand side,
substituting our names:
returnState a bindState f
==>
\st -> let (st, a) = (returnState a) st
m
= f a
in m st
==>
\st -> let (st, a) = (\st -> (st, a)) st
in (f a) st
==>
\st -> let (st, a) = (st, a)
in (f a) st
127
==>
\st -> (f a) st
==>
f a
In the first step, we simply substitute the definition of bindState. In the second
step, we simplify the last two lines and substitute the definition of returnState. In
the third step, we apply st to the lambda function. In the fourth step, we rename st
to st and remove the let. In the last step, we eta reduce.
Moving on to Law 2, we need to show that f >>= return f. This is shown
as follows:
f bindState returnState
==>
\st -> let (st, a) = f st
in (returnState a) st
==>
\st -> let (st, a) = f st
in (\st -> (st, a)) st
==>
\st -> let (st, a) = f st
in (st, a)
==>
\st -> f st
==>
f
Finally, we need to show that State obeys the third law: f >>= (\x -> g x
>>= h) (f >>= g) >>= h. This is much more involved to show, so we will
only sketch the proof here. Notice that we can write the left-hand-side as:
\st -> let (st, a) = f st
in (\x -> g x bindState h) a st
==>
\st -> let (st, a) = f st
in (g a bindState h) st
==>
\st -> let (st, a) = f st
in (\st -> let (st, b) = g a
in h b st) st
==>
\st -> let (st , a) = f st
(st, b) = g a st
(st,c) = h b st
in (st,c)
CHAPTER 9. MONADS
128
The interesting thing to note here is that we have both action applications on the
same let level. Since let is associative, this means that we can put whichever bracketing
we prefer and the results will not change. Of course, this is an informal, hand waving
argument and it would take us a few more derivations to actually prove, but this gives
the general idea.
Now that we know that State st is actually a monad, wed like to make it an
instance of the Monad class. Unfortunately, the straightforward way of doing this
doesnt work. We cant write:
instance Monad (State st) where { ... }
This is because you cannot make instances out of non-fully-applied type synonyms.
Instead, what we need to do instead is convert the type synonym into a newtype, as:
newtype State st a = State (st -> (st, a))
Unfortunately, this means that we need to do some packing and unpacking of the
State constructor in the Monad instance declaration, but its not terribly difficult:
instance Monad (State state) where
return a = State (\state -> (state, a))
State run >>= action = State run
where run st =
let (st, a)
= run st
State run = action a
in run st
mapTreeM
129
That is, mapTreeM can be run in any monad, not just our State monad.
Now, the nice thing about encapsulating the stateful aspect of the computation like
this is that we can provide functions to get and change the current state. These look
like:
getState :: State state state
getState = State (\state -> (state, state))
putState :: state -> State state ()
putState new = State (\_ -> (new, ()))
Here, getState is a monadic operation that takes the current state, passes it
through unchanged, and then returns it as the value. The putState function takes a
new state and produces an action that ignores the current state and inserts the new one.
Now, we can write our numberTree function as:
numberTree :: Tree a -> State Int (Tree (a, Int))
numberTree tree = mapTreeM number tree
where number v = do
cur <- getState
putState (cur+1)
return (v,cur)
Finally, we need to be able to run the action by providing an initial state:
runStateM :: State state a -> state -> a
runStateM (State f) st = snd (f st)
Now, we can provide an example Tree:
testTree =
Branch
(Branch
(Leaf a)
(Branch
(Leaf b)
(Leaf c)))
(Branch
(Leaf d)
(Leaf e))
and number it:
State> runStateM (numberTree testTree) 1
Branch (Branch (Leaf (a,1)) (Branch (Leaf (b,2))
(Leaf (c,3)))) (Branch (Leaf (d,4))
(Leaf (e,5)))
getState
putState
CHAPTER 9. MONADS
130
This may seem like a large amount of work to do something simple. However,
note the new power of mapTreeM. We can also print out the leaves of the tree in a
left-to-right fashion as:
State> mapTreeM print testTree
a
b
c
d
e
This crucially relies on the fact that mapTreeM has the more general type involving
arbitrary monads not just the state monad. Furthermore, we can write an action that
will make each leaf value equal to its old value as well as all the values preceeding:
fluffLeaves tree = mapTreeM fluff tree
where fluff v = do
cur <- getState
putState (v:cur)
return (v:cur)
and can see it in action:
State> runStateM (fluffLeaves testTree) []
Branch (Branch (Leaf "a") (Branch (Leaf "ba")
(Leaf "cba"))) (Branch (Leaf "dcba")
(Leaf "edcba"))
In fact, you dont even need to write your own monad instance and datatype. All
this is built in to the Control.Monad.State module. There, our runStateM
is called evalState; our getState is called get; and our putState is called
put.
This module also contains a state transformer monad, which we will discuss in
Section 9.7.
It turns out that many of our favorite datatypes are actually monads themselves. Consider, for instance, lists. They have a monad definition that looks something like:
instance Monad
return x =
l >>= f =
fail _
=
[] where
[x]
concatMap f l
[]
131
This enables us to use lists in do notation. For instance, given the definition:
cross l1 l2 = do
x <- l1
y <- l2
return (x,y)
we get a cross-product function:
Monads> cross "ab" "cde"
[(a,d),(a,e),(a,f),(b,d),(b,e),
(b,f)]
It is not a coincidence that this looks very much like the list comprehension form:
list comprehensions
Maybe
132
CHAPTER 9. MONADS
What this means is that if we write a function (like searchAll from Section 8.4)
only in terms of monadic operators, we can use it with any monad, depending on what
we mean. Using real monadic functions (not do notation), the searchAll function
looks something like:
searchAll g@(Graph vl el) src dst
| src == dst = return [src]
| otherwise = search el
where search [] = fail "no path"
search ((u,v,_):es)
| src == u =
searchAll g v dst >>= \path ->
return (u:path)
| otherwise = search es
The type of this function is Monad m => Graph v e -> Int -> Int ->
m [Int]. This means that no matter what monad were using at the moment, this
function will perform the calculation. Suppose we have the following graph:
gr = Graph [(0, a), (1, b), (2, c), (3, d)]
[(0,1,l), (0,2,m), (1,3,n), (2,3,m)]
This represents a graph with four nodes, labelled a,b,c and d. There is an edge
from a to both b and c. There is also an edge from both b and c to d. Using the Maybe
monad, we can compute the path from a to d:
Monads> searchAll gr 0 3 :: Maybe [Int]
Just [0,1,3]
We provide the type signature, so that the interpreter knows what monad were
using. If we try to search in the opposite direction, there is no path. The inability to
find a path is represented as Nothing in the Maybe monad:
Monads> searchAll gr 3 0 :: Maybe [Int]
Nothing
Note that the string no path has disappeared since theres no way for the Maybe
monad to record this.
If we perform the same impossible search in the list monad, we get the empty list,
indicating no path:
Monads> searchAll gr 3 0 :: [[Int]]
[]
sea
133
If we perform the possible search, we get back a list containing the first path:
Monads> searchAll gr 0 3 :: [[Int]]
[[0,1,3]]
You may have expected this function call to return all paths, but, as coded, it does
not. See Section 9.6 for more about using lists to represent nondeterminism.
If we use the IO monad, we can actually get at the error message, since IO knows
how to keep track of error messages:
Monads> searchAll gr 0 3 :: IO [Int]
Monads> it
[0,1,3]
Monads> searchAll gr 3 0 :: IO [Int]
*** Exception: user error
Reason: no path
In the first case, we needed to type it to get GHCi to actually evaluate the search.
There is one problem with this implementation of searchAll: if it finds an edge
that does not lead to a solution, it wont be able to backtrack. This has to do with
the recursive call to searchAll inside of search. Consider, for instance, what
happens if searchAll g v dst doesnt find a path. Theres no way for this implementation to recover. For instance, if we remove the edge from node b to node d,
we should still be able to find a path from a to d, but this algorithm cant find it. We
define:
gr2 = Graph [(0, a), (1, b), (2, c), (3, d)]
[(0,1,l), (0,2,m), (2,3,m)]
and then try to search:
Monads> searchAll gr2 0 3
*** Exception: user error
Reason: no path
To fix this, we need a function like combine from our Computation class. We
will see how to do this in Section 9.6.
Exercises
Exercise 9.1 Verify that Maybe obeys the three monad laws.
Exercise 9.2 The type Either String is a monad that can keep track of errors. Write an
instance for it, and then try doing the search from this chapter using this monad.
Hint: Your instance declaration should begin: instance Monad (Either String)
where.
nondeterminism
CHAPTER 9. MONADS
134
mapM ::
mapM ::
filterM ::
foldM ::
sequence ::
[m a] -> m [a]
sequence ::
[m a] -> m ()
liftM ::
when ::
join ::
m (m a) -> m a
The mapM, filterM and foldM are our old friends map, filter and foldr
wrapped up inside of monads. These functions are incredibly useful (particularly
foldM) when working with monads. We can use mapM , for instance, to print a list of
things to the screen:
Monads> mapM_ print [1,2,3,4,5]
1
2
3
4
5
135
We can use foldM to sum a list and print the intermediate sum at each step:
Monads> foldM (\a b ->
putStrLn (show a ++ "+" ++ show b ++
"=" ++ show (a+b)) >>
return (a+b)) 0 [1..5]
0+1=1
1+2=3
3+3=6
6+4=10
10+5=15
Monads> it
15
The sequence and sequence functions simply execute a list of actions. For
instance:
sequence
liftM
CHAPTER 9. MONADS
136
The when function executes a monadic action only if a condition is met. So, if we
only want to print non-empty lines:
Monads> mapM_ (\l -> when (not $ null l) (putStrLn l))
["","abc","def","","","ghi"]
abc
def
ghi
join
Of course, the same could be accomplished with filter, but sometimes when is
more convenient.
Finally, the join function is the monadic equivalent of concat on lists. In fact,
when m is the list monad, join is exactly concat. In other monads, it accomplishes
a similar task:
Monads> join (Just (Just a))
Just a
Monads> join (Just (Nothing :: Maybe Char))
Nothing
Monads> join (Nothing :: Maybe (Maybe Char))
Nothing
9.6. MONADPLUS
137
9.6 MonadPlus
Given only the >>= and return functions, it is impossible to write a function like
combine with type c a c a c a. However, such a function is so generally useful
that it exists in another class called MonadPlus. In addition to having a combine
function, instances of MonadPlus also have a zero element that is the identity
under the plus (i.e., combine) action. The definition is:
combine
MonadPlus
Maybe
lists
CHAPTER 9. MONADS
138
| otherwise = search el
where search [] = fail "no path"
search ((u,v,_):es)
| src == u =
(searchAll2 g v dst >>= \path ->
return (u:path)) mplus
search es
| otherwise = search es
Now, when were going through the edge list in search, and we come across a
matching edge, not only do we explore this path, but we also continue to explore the
out-edges of the current node in the recursive call to search.
The IO monad is not an instance of MonadPlus; we were not able to execute
the search with this monad. We can see that when using lists as the monad, we (a) get
all possible paths in gr and (b) get a path in gr2.
MPlus> searchAll2 gr 0 3 :: [[Int]]
[[0,1,3],[0,2,3]]
MPlus> searchAll2 gr2 0 3 :: [[Int]]
[[0,2,3]]
You might be tempted to implement this as:
searchAll2 g@(Graph vl el) src dst
| src == dst = return [src]
| otherwise = search el
where search [] = fail "no path"
search ((u,v,_):es)
| src == u = do
path <- searchAll2 g v dst
rest <- search es
return ((u:path) mplus rest)
| otherwise = search es
But note that this doesnt do what we want. Here, if the recursive call to searchAll2
fails, we dont try to continue and execute search es. The call to mplus must be
at the top level in order for it to work.
Exercises
Exercise 9.3 Suppose that we changed the order of arguments to mplus. I.e., the
matching case of search looked like:
search es mplus
(searchAll2 g v dst >>= \path ->
return (u:path))
139
How would you expect this to change the results when using the list monad on gr?
Why?
MonadTrans
lift
state monad
state transformer
140
CHAPTER 9. MONADS
getT
putT
evalStateT
The idea behind the definition of return is that we keep the state constant and
simply return the state/a pair in the enclosed monad. Note that the use of return in
the definition of return refers to the enclosed monad, not the state transformer.
In the definition of bind, we create a new StateT that takes a state s as an argument. First, it applies this state to the first action (StateT m) and gets the new
state and answer as a result. It then runs the k action on this new state and gets a new
transformer. It finally applies the new state to this transformer. This definition is nearly
identical to the definition of bind for the standard (non-transformer) State monad
described in Section 9.3.
The fail function passes on the call to fail in the enclosed monad, since state
transformers dont natively know how to deal with failure.
Of course, in order to actually use this monad, we need to provide function getT
, putT and evalStateT . These are analogous to getState, putState and
runStateM from Section 9.3:
getT :: Monad m => StateT s m s
getT = StateT (\s -> return (s, s))
putT :: Monad m => s -> StateT s m ()
putT s = StateT (\_ -> return (s, ()))
evalStateT :: Monad m => StateT s m a -> s -> m a
evalStateT (StateT m) state = do
(s, a) <- m state
return a
These functions should be straightforward. Note, however, that the result of evalStateT
is actually a monadic action in the enclosed monad. This is typical of monad transformers: they dont know how to actually run things in their enclosed monad (they
only know how to lift actions). Thus, what you get out is a monadic action in the
inside monad (in our case, IO), which you then need to run yourself.
We can use state transformers to reimplement a version of our mapTreeM function
from Section 9.3. The only change here is that when we get to a leaf, we print out the
value of the leaf; when we get to a branch, we just print out Branch.
mapTreeM action (Leaf a) = do
lift (putStrLn ("Leaf " ++ show a))
141
b <- action a
return (Leaf b)
mapTreeM action (Branch lhs rhs) = do
lift (putStrLn "Branch")
lhs <- mapTreeM action lhs
rhs <- mapTreeM action rhs
return (Branch lhs rhs)
The only difference between this function and the one from Section 9.3 is the calls
to lift (putStrLn ...) as the first line. The lift tells us that were going to
be executing a command in an enclosed monad. In this case, the enclosed monad is
IO, since the command lifted is putStrLn.
The type of this function is relatively complex:
mapTreeM :: (MonadTrans t, Monad (t IO), Show a) =>
(a -> t IO a1) -> Tree a -> t IO (Tree a1)
Ignoring, for a second, the class constraints, this says that mapTreeM takes an
action and a tree and returns a tree. This just as before. In this, we require that t is
a monad transformer (since we apply lift in it); we require that t IO is a monad,
since we use putStrLn we know that the enclosed monad is IO; finally, we require
that a is an instance of show this is simply because we use show to show the value
of leaves.
Now, we simply change numberTree to use this version of mapTreeM, and the
new versions of get and put, and we end up with:
numberTree tree = mapTreeM number tree
where number v = do
cur <- getT
putT (cur+1)
return (v,cur)
Using this, we can run our monad:
MTrans> evalStateT (numberTree testTree) 0
Branch
Branch
Leaf a
Branch
Leaf b
Leaf c
Branch
Leaf d
Leaf e
*MTrans> it
142
CHAPTER 9. MONADS
cycles
One problem not specified in our discussion of MonadPlus is that our search
algorithm will fail to terminate on graphs with cycles. Consider:
gr3 = Graph [(0, a), (1, b), (2, c), (3, d)]
[(0,1,l), (1,0,m), (0,2,n),
(1,3,o), (2,3,p)]
In this graph, there is a back edge from node b back to node a. If we attempt to run
searchAll2, regardless of what monad we use, it will fail to terminate. Moreover,
if we move this erroneous edge to the end of the list (and call this gr4), the result of
searchAll2 gr4 0 3 will contain an infinite number of paths: presumably we
only want paths that dont contain cycles.
In order to get around this problem, we need to introduce state. Namely, we need
to keep track of which nodes we have visited, so that we dont visit them again.
We can do this as follows:
searchAll5 g@(Graph vl el) src dst
| src == dst = do
visited <- getT
putT (src:visited)
return [src]
| otherwise = do
visited <- getT
putT (src:visited)
if src elem visited
then mzero
else search el
where
search [] = mzero
search ((u,v,_):es)
| src == u =
(do path <- searchAll5 g v dst
return (u:path)) mplus
search es
| otherwise = search es
Here, we implicitly use a state transformer (see the calls to getT and putT) to
keep track of visited states. We only continue to recurse, when we encounter a state we
havent yet visited. Futhermore, when we recurse, we add the current state to our set
of visited states.
Now, we can run the state transformer and get out only the correct paths, even on
the cyclic graphs:
143
Exercises
Exercise 9.4 Write a function searchAll6, based on the code for searchAll2,
that, at every entry to the main function (not the recursion over the edge list), prints the
search being conducted. For instance, the output generated for searchAll6 gr 0
3 should look like:
Exploring 0 -> 3
Exploring 1 -> 3
Exploring 3 -> 3
Exploring 2 -> 3
Exploring 3 -> 3
MTrans> it
[[0,1,3],[0,2,3]]
In order to do this, you will have to define your own list monad transformer and make
appropriate instances of it.
Exercise 9.5 Combine the searchAll5 function (from this section) with the searchAll6
function (from the previous exercise) into a single function called searchAll7. This
function should perform IO as in searchAll6 but should also keep track of state
using a state transformer.
CHAPTER 9. MONADS
144
9.8.1
Consider the task of parsing. A simple parsing monad is much like a state monad,
where the state is the unparsed string. We can represent this exactly as:
newtype Parser a = Parser
{ runParser :: String -> Either String (String, a) }
We again use Left err to be an error condition. This yields standard instances
of Monad and MonadPlus:
instance Monad Parser where
return a = Parser (\xl -> Right (xl,a))
fail
s = Parser (\xl -> Left s)
Parser m >>= k = Parser $ \xl ->
case m xl of
Left s -> Left s
Right (xl, a) ->
let Parser n = k a
in n xl
instance MonadPlus Parser where
mzero = Parser (\xl -> Left "mzero")
Parser p mplus Parser q = Parser $ \xl ->
case p xl of
Right a -> Right a
Left err -> case q xl of
Right a -> Right a
Left _ -> Left err
primitives
Now, we want to build up a library of paring primitives. The most basic primitive
is a parser that will read a specific character. This function looks like:
char :: Char -> Parser Char
char c = Parser char
where char [] = Left ("expecting " ++ show c ++
" got EOF")
char (x:xs)
| x == c
= Right (xs, c)
145
("expecting " ++
show c ++ " got " ++
show x)
Here, the parser succeeds only if the first character of the input is the expected
character.
We can use this parser to build up a parser for the string Hello:
helloParser :: Parser String
helloParser = do
char H
char e
char l
char l
char o
return "Hello"
This shows how easy it is to combine these parsers. We dont need to worry about
the underlying string the monad takes care of that for us. All we need to do is
combine these parser primatives. We can test this parser by using runParser and by
supplying input:
Parsing> runParser helloParser "Hello"
Right ("","Hello")
Parsing> runParser helloParser "Hello World!"
Right (" World!","Hello")
Parsing> runParser helloParser "hello World!"
Left "expecting H got h"
We can have a slightly more general function, which will match any character fitting
a description:
matchChar :: (Char -> Bool) -> Parser Char
matchChar c = Parser matchChar
where matchChar [] =
Left ("expecting char, got EOF")
matchChar (x:xs)
| c x
= Right (xs, x)
| otherwise =
Left ("expecting char, got " ++
show x)
Using this, we can write a case-insensitive Hello parser:
runParser
CHAPTER 9. MONADS
146
ciHelloParser = do
c1 <- matchChar (elem
c2 <- matchChar (elem
c3 <- matchChar (elem
c4 <- matchChar (elem
c5 <- matchChar (elem
return [c1,c2,c3,c4,c5]
"Hh")
"Ee")
"Ll")
"Ll")
"Oo")
toLower),
On top of these primitives, we usually build some combinators. The many combinator, for instance, will take a parser that parses entities of type a and will make it into
a parser that parses entities of type [a] (this is a Kleene-star operator):
many :: Parser a -> Parser [a]
many (Parser p) = Parser many
where many xl =
case p xl of
Left err -> Right (xl, [])
Right (xl,a) ->
let Right (xl, rest) = many xl
in Right (xl, a:rest)
The idea here is that first we try to apply the given parser, p. If this fails, we succeed
but return the empty list. If p succeeds, we recurse and keep trying to apply p until it
fails. We then return the list of successes weve accumulated.
In general, there would be many more functions of this sort, and they would be hidden away in a library, so that users couldnt actually look inside the Parser type.
However, using them, you could build up, for instance, a parser that parses (nonnegative) integers:
147
mplus
CHAPTER 9. MONADS
148
tricky =
mplus (string "Hal") (string "Hall")
You might expect this parser to parse both the words Hal and Hall; however, it
only parses the former. You can see this with:
Parsing> runParser tricky "Hal"
Right ("","Hal")
Parsing> runParser tricky "Hall"
Right ("l","Hal")
This is because it tries to parse Hal, which succeeds, and then it doesnt bother
trying to parse Hall.
You can attempt to fix this by providing a parser primitive, which detects end-of-file
(really, end-of-string) as:
eof :: Parser ()
eof = Parser eof
where eof [] = Right ([], ())
eof xl = Left ("Expecting EOF, got " ++
show (take 10 xl))
You might then rewrite tricky using eof as:
tricky2 = do
s <- mplus (string "Hal") (string "Hall")
eof
return s
But this also doesnt work, as we can easily see:
Parsing> runParser tricky2 "Hal"
Right ("",())
Parsing> runParser tricky2 "Hall"
Left "Expecting EOF, got \"l\""
This is because, again, the mplus doesnt know that it needs to parse the whole
input. So, when you provide it with Hall, it parses just Hal and leaves the last l
lying around to be parsed later. This causes eof to produce an error message.
The correct way to implement this is:
tricky3 =
mplus (do s <- string "Hal"
149
eof
return s)
(do s <- string "Hall"
eof
return s)
We can see that this works:
Parsing> runParser tricky3 "Hal"
Right ("","Hal")
Parsing> runParser tricky3 "Hall"
Right ("","Hall")
This works precisely because each side of the mplus knows that it must read the
end.
In this case, fixing the parser to accept both Hal and Hall was fairly simple,
due to the fact that we assumed we would be reading an end-of-file immediately afterwards. Unfortunately, if we cannot disambiguate immediately, life becomes significantly more complicated. This is a general problem in parsing, and has little to do with
monadic parsing. The solution most parser libraries (e.g., Parsec, see Section 9.8.2)
have adopted is to only recognize LL(1) grammars: that means that you must be able
to disambiguate the input with a one token look-ahead.
Exercises
Exercise 9.6 Write a parser intListSpace that will parse int lists but will allow
arbitrary white space (spaces, tabs or newlines) between the commas and brackets.
Given this monadic parser, it is fairly easy to add information regarding source
position. For instance, if were parsing a large file, it might be helpful to report the
line number on which an error occurred. We could do this simply by extending the
Parser type and by modifying the instances and the primitives:
newtype Parser a = Parser
{ runParser :: Int -> String ->
Either String (Int, String, a) }
instance Monad Parser where
return a = Parser (\n xl -> Right (n,xl,a))
fail
s = Parser (\n xl -> Left (show n ++
": " ++ s))
Parser m >>= k = Parser $ \n xl ->
case m n xl of
Left s -> Left s
Right (n, xl, a) ->
let Parser m2 = k a
line numbers
CHAPTER 9. MONADS
150
in
m2 n xl
151
9.8.2
Parsec
As you continue developing your parser, you might want to add more and more features. Luckily, Graham Hutton and Daan Leijen have already done this for us in the
Parsec library. This section is intended to be an introduction to the Parsec library; it by
no means covers the whole library, but it should be enough to get you started.
Like our libarary, Parsec provides a few basic functions to build parsers from characters. These are: char, which is the same as our char; anyChar, which is the same
as our anyChar; satisfy, which is the same as our matchChar; oneOf, which
takes a list of Chars and matches any of them; and noneOf, which is the opposite of
oneOf.
The primary function Parsec uses to run a parser is parse. However, in addition to
a parser, this function takes a string that represents the name of the file youre parsing.
This is so it can give better error messages. We can try parsing with the above functions:
char
anyChar
satisfy
oneOf
noneOf
parse
spaces
space
letter
digit
string
CHAPTER 9. MONADS
152
153
choice takes a list of parsers and performs an or operation (<|>) between all
of them.
option takes a default value of type a and a parser that returns something of
type a. It then tries to parse with the parser, but it uses the default value as the
return, if the parsing fails.
optional takes a parser that returns () and optionally runs it.
between takes three parsers: an open parser, a close parser and a between
parser. It runs them in order and returns the value of the between parser. This
can be used, for instance, to take care of the brackets on our intList parser.
notFollowedBy takes a parser and returns one that succeeds only if the given
parser would have failed.
Suppose we want to parse a simple calculator language that includes only plus
and times. Furthermore, for simplicity, assume each embedded expression must be
enclosed in parentheses. We can give a datatype for this language as:
data Expr = Value Int
| Expr :+: Expr
| Expr :*: Expr
deriving (Eq, Ord, Show)
And then write a parser for this language as:
parseExpr :: Parser Expr
parseExpr = choice
[ do i <- int; return (Value i)
, between (char () (char )) $ do
e1 <- parseExpr
op <- oneOf "+*"
e2 <- parseExpr
case op of
+ -> return (e1 :+: e2)
* -> return (e1 :*: e2)
]
Here, the parser alternates between two options (we could have used <|>, but I
wanted to show the choice combinator in action). The first simply parses an int and
then wraps it up in the Value constructor. The second option uses between to parse
text between parentheses. What it parses is first an expression, then one of plus or
times, then another expression. Depending on what the operator is, it returns either e1
:+: e2 or e1 :*: e2.
We can modify this parser, so that instead of computing an Expr, it simply computes the value:
CHAPTER 9. MONADS
154
Now, suppose we want to introduce bindings into our language. That is, we want
to also be able to say let x = 5 in inside of our expressions and then use the variables
weve defined. In order to do this, we need to use the getState and setState (or
updateState) functions built in to Parsec.
parseValueLet :: CharParser (FiniteMap Char Int) Int
parseValueLet = choice
[ int
, do string "let "
c <- letter
char =
e <- parseValueLet
string " in "
updateState (\fm -> addToFM fm c e)
parseValueLet
, do c <- letter
fm <- getState
case lookupFM fm c of
Nothing -> unexpected ("variable " ++ show c ++
" unbound")
Just i -> return i
, between (char () (char )) $ do
e1 <- parseValueLet
op <- oneOf "+*"
e2 <- parseValueLet
case op of
+ -> return (e1 + e2)
155
Exercises
Exercise 9.7 Modify the parseValueLet parser, so that it obeys bracketing. In
order to do this, you will need to change the state to something like FiniteMap
Char [Int], where the [Int] is a stack of definitions.
runParser
156
CHAPTER 9. MONADS
Chapter 10
Advanced Techniques
10.1 Exceptions
10.2 Mutable Arrays
10.3 Mutable References
10.4 The ST Monad
10.5 Concurrency
10.6 Regular Expressions
10.7 Dynamic Types
157
158
Appendix A
160
want to consider them all that important. Therefore, we say that the complexity of this
sum function is O(n) (read order n). Basically saying something is O(n) means that
for some constant factors x and y, the function takes nx+y machine steps to complete.
Consider the following sorting algorithm for lists (commonly called insertion
sort):
sort [] = []
sort [x] = [x]
sort (x:xs) = insert (sort xs)
where insert [] = [x]
insert (y:ys) | x <= y
= x : y : ys
| otherwise = y : insert ys
The way this algorithm works is as follow: if we want to sort an empty list or a list
of just one element, we return them as they are, as they are already sorted. Otherwise,
we have a list of the form x:xs. In this case, we sort xs and then want to insert x
in the appropriate location. Thats what the insert function does. It traverses the
now-sorted tail and inserts x wherever it naturally fits.
Lets analyze how long this function takes to complete. Suppose it takes f (n) stepts
to sort a list of length n. Then, in order to sort a list of n-many elements, we first have
to sort the tail of the list first, which takes f (n 1) time. Then, we have to insert x into
this new list. If x has to go at the end, this will take O(n 1) = O(n) steps. Putting
all of this together, we see that we have to do O(n) amount of work O(n) many times,
which means that the entire complexity of this sorting algorithm is O(n2 ). Here, the
squared is not a constant value, so we cannot throw it out.
What does this mean? Simply that for really long lists, the sum function wont take
very long, but that the sort function will take quite some time. Of course there are
algorithms that run much more slowly that simply O(n2 ) and there are ones that run
more quickly than O(n).
Consider the random access functions for lists and arrays. In the worst case, accessing an arbitrary element in a list of length n will take O(n) time (think about
accessing the last element). However with arrays, you can access any element immediately, which is said to be in constant time, or O(1), which is basically as fast an any
algorithm can go.
Theres much more in complexity theory than this, but this should be enough to
allow you to understand all the discussions in this tutorial. Just keep in mind that O(1)
is faster than O(n) is faster than O(n2 ), etc.
Appendix B
162
There is a close connection between recursion and mathematical induction. Induction is a proof technique which typically breaks problems down into base cases and
inductive cases, very analogous to our analysis of recursion.
Lets say we want to prove the statement n! n for all n 0. First we formulate
a base case: namely, we wish to prove the statement when n = 0. When n = 0, n! = 1
by definition. Since n! = 1 > 0 = n, we get that 0! 0 as desired.
Now, suppose that n > 0. Then n = k + 1 for some value k. We now invoke the
inductive hypothesis and claim that the statement holds for n = k. That is, we assume
that k! k. Now, we use k to formate the statement for our value of n. That is, n! n
if and only iff (k + 1)! (k + 1). We now apply the definition of factorial and get
(k + 1)! = (k + 1) k!. Now, we know k! k, so (k + 1) k! k + 1 if and only if
k + 1 1. But we know that k 0, which means k + 1 1. Thus it is proven.
It may seem a bit counter-intuitive that we are assuming that the claim is true for k
in our proof that it is true for n. You can think of it like this: weve proved the statement
for the case when n = 0. Now, we know its true for n = 0 so using this we use our
inductive argument to show that its true for n = 1. Now, we know that it is true for
n = 1 so we reuse our inductive argument to show that its true for n = 2. We can
continue this argument as long as we want and then see that its true for all n.
Its much like pushing down dominoes. You know that when you push down the
first domino, its going to knock over the second one. This, in turn will knock over the
third, and so on. The base case is like pushing down the first domino, and the inductive
case is like showing that pushing down domino k will cause the k + 1st domino to fall.
In fact, we can use induction to prove that our natToInt function does the right
thing. First we prove the base case: does natToInt Zero evaluate to 0? Yes, obviously it does. Now, we can assume that natToInt n evaluates to the correct value
(this is the inductive hypothesis) and ask whether natToInt (Succ n) produces
the correct value. Again, it is obvious that it does, by simply looking at the definition.
Lets consider a more complex example: addition of Nats. We can write this concisely as:
addNat Zero m = m
addNat (Succ n) m = addNat n (Succ m)
Now, lets prove that this does the correct thing. First, as the base case, suppose the
first argument is Zero. We know that 0 + m = m regardless of what m is; thus in the
base case the algorithm does the correct thing. Now, suppose that addNat n m does
the correct thing for all m and we want to show that addNat (Succ n) m does the
correct thing. We know that (n + 1) + m = n + (m + 1) and thus since addNat
n (Succ m) does the correct thing (by the inductive hypothesis), our program is
correct.
Appendix C
Solutions To Exercises
Solution 3.1
It binds more tightly; actually, function application binds more tightly than anything
else. To see this, we can do something like:
Prelude> sqrt 3 * 3
5.19615
If multiplication bound more tightly, the result would have been 3.
Solution 3.2
Solution: snd (fst ((1,a),"foo")). This is because first we want to take
the first half the the tuple: (1,a) and then out of this we want to take the second
half, yielding just a.
If you tried fst (snd ((1,a),"foo")) you will have gotten a type error.
This is because the application of snd will leave you with fst "foo". However, the
string foo isnt a tuple, so you cannot apply fst to it.
Solution 3.3
Solution: map Char.isLower aBCde
Solution 3.4
Solution: length (filter Char.isLower aBCde)
Solution 3.5
foldr max 0 [5,10,2,8,1]. You could also use foldl. The foldr case is easier to explain:
we replace each cons with an application of max and the empty list with 0. Thus, the
inner-most application will take the maximum of 0 and the last element of the list (if
it exists). Then, the next-most inner application will return the maximum of whatever was the maximum before and the second-to-last element. This will continue on,
carrying to current maximum all the way back to the beginning of the list.
163
164
In the foldl case, we can think of this as looking at each element in the list in order.
We start off our state with 0. We pull off the first element and check to see if its
bigger than our current state. If it is, we replace our current state with that number and
the continue. This happens for each element and thus eventually returns the maximal
element.
Solution 3.6
fst (head (tail [(5,b),(1,c),(6,a)]))
Solution 3.7
We can define a fibonacci function as:
fib 1 = 1
fib 2 = 1
fib n = fib (n-1) + fib (n-2)
We could also write it using explicit if statements, like:
fib n =
if n == 1 || n == 2
then 1
else fib (n-1) + fib (n-2)
Either is acceptable, but the first is perhaps more natural in Haskell.
Solution 3.8
We can define:
ab=
a
a + a (b 1)
b=1
otherwise
Solution 3.9
We can define my map as:
165
my_map f [] = []
my_map f (x:xs) = f x : my_map f xs
Recall that the my map function is supposed to apply a function f to every element
in the list. In the case that the list is empty, there are no elements to apply the function
to, so we just return the empty list.
In the case that the list is non-empty, it is an element x followed by a list xs.
Assuming weve already properly applied my map to xs, then all were left to do is
apply f to x and then stick the results together. This is exactly what the second line
does.
Solution 3.10
The code below appears in Numbers.hs. The only tricky parts are the recursive calls
in getNums and showFactorials.
module Main
where
import IO
main = do
nums <- getNums
putStrLn ("The sum is " ++ show (sum nums))
putStrLn ("The product is " ++ show (product nums))
showFactorials nums
getNums = do
putStrLn "Give me a number (or 0 to stop):"
num <- getLine
if read num == 0
then return []
else do rest <- getNums
return ((read num :: Int):rest)
showFactorials []
= return ()
showFactorials (x:xs) = do
putStrLn (show x ++ " factorial is " ++
show (factorial x))
showFactorials xs
factorial 1 = 1
factorial n = n * factorial (n-1)
The idea for getNums is just as spelled out in the hint. For showFactorials,
we consider first the recursive call. Suppose we have a list of numbers, the first of
166
which is x. First we print out the string showing the factorial. Then we print out the
rest, hence the recursive call. But what should we do in the case of the empty list?
Clearly we are done, so we dont need to do anything at all, so we simply return
().
Note that this must be return () instead of just () because if we simply wrote
showFactorials [] = () then this wouldnt be an IO action, as it needs to be.
For more clarification on this, you should probably just keep reading the tutorial.
Solution 4.1
String or [Char]
2. type error: lists are homogenous
3. Num a (a, Char)
4. Int
5. type error: cannot add values of different types
Solution 4.2
The types:
1. (a, b) > b
2. [a] > a
3. [a] > Bool
4. [a] > a
5. [[a]] > a
Solution 4.3
The types:
1. a > [a]. This function takes an element and returns the list containing only that
element.
2. a > b > b > (a, [b]). The second and third argument must be of the same
type, since they go into the same list. The first element can be of any type.
3. Num a => a > a. Since we apply (+) to a, it must be an instance of Num.
4. a > String. This ignores the first argument, so it can be any type.
5. (Char > a) > a. In this expression, x must be a function which takes a Char
as an argument. We dont know anything about what it produces, though, so we
call it a.
1.
167
6. Type error. Here, we assume x has type a. But x is applied to itself, so it must
have type b > c. But then it must have type (b > c) > c, but then it must
have type ((b > c) > c) > c and so on, leading to an infinite type.
7. Num a => a > a. Again, since we apply (+), this must be an instance of
Num.
Solution 4.4
The definitions will be something like:
data Triple a b c = Triple a b c
tripleFst (Triple x y z) = x
tripleSnd (Triple x y z) = y
tripleThr (Triple x y z) = z
Solution 4.5
The code, with type signatures, is:
data Quadruple a b = Quadruple a a b b
firstTwo :: Quadruple a b -> [a]
firstTwo (Quadruple x y z t) = [x,y]
lastTwo :: Quadruple a b -> [b]
lastTwo (Quadruple x y z t) = [z,t]
We note here that there are only two type variables, a and b associated with
Quadruple.
Solution 4.6
The code:
data Tuple a b c d e =
|
|
|
One a
Two a b
Three a b c
Four a b c d
tuple1
tuple1
tuple1
tuple1
=
=
=
=
(One
(Two
(Three
(Four
a
)
a b
)
a b c )
a b c d)
tuple2 (One
a
tuple2 (Two
a b
tuple2 (Three a b c
Just
Just
Just
Just
a
a
a
a
) = Nothing
) = Just b
) = Just b
168
tuple2 (Four
a b c d) = Just b
tuple3
tuple3
tuple3
tuple3
(One
(Two
(Three
(Four
a
)
a b
)
a b c )
a b c d)
=
=
=
=
Nothing
Nothing
Just c
Just c
tuple4
tuple4
tuple4
tuple4
(One
(Two
(Three
(Four
a
)
a b
)
a b c )
a b c d)
=
=
=
=
Nothing
Nothing
Nothing
Just d
Solution 4.7
The code:
fromTuple
fromTuple
fromTuple
fromTuple
(One
(Two
(Three
(Four
a
)
a b
)
a b c )
a b c d)
=
=
=
=
Left
Left
Right
Right
(Left
(Right
(Left
(Right
a
)
(a,b)
)
(a,b,c) )
(a,b,c,d))
Here, we use embedded Eithers to represent the fact that there are four (instead
of two) options.
Solution 4.8
The code:
listHead (Cons x xs) = x
listTail (Cons x xs) = xs
listFoldl f y Nil = y
listFoldl f y (Cons x xs) = listFoldl f (f y x) xs
listFoldr f y Nil = y
listFoldr f y (Cons x xs) = f x (foldr f z xs)
Solution 4.9
The code:
elements (Leaf x) = [x]
elements (Branch lhs x rhs) =
elements lhs ++ [x] ++ elements rhs
Solution 4.10
The code:
169
Solution 4.11
It mimicks neither exactly. Its behavior most closely resembles foldr, but differs
slightly in its treatment of the initial value. We can observe the difference in an interpreter:
CPS> foldr (-) 0 [1,2,3]
2
CPS> foldl (-) 0 [1,2,3]
-6
CPS> fold (-) 0 [1,2,3]
-2
Clearly it behaves differently. By writing down the derivations of fold and foldr
we can see exactly where they diverge:
==>
==>
==>
==>
==>
==>
==>
==>
==>
==>
Essentially, the primary difference is that in the foldr case, the initial value is
used at the end (replacing []), whereas in the CPS case, the initial value is used at the
beginning.
170
Solution 4.12
Solution 5.1
Using if, we get something like:
main = do
putStrLn "Please enter your name:"
name <- getLine
if name == "Simon" || name == "John" || name == "Phil"
then putStrLn "Haskell is great!"
else if name == "Koen"
then putStrLn "Debugging Haskell is fun!"
else putStrLn "I dont know who you are."
Note that we dont need to repeat the dos inside the ifs, since these are only one
action commands.
We could also be a bit smarter and use the elem command which is built in to the
Prelude:
main = do
putStrLn "Please enter your name:"
name <- getLine
if name elem ["Simon", "John", "Phil"]
then putStrLn "Haskell is great!"
else if name == "Koen"
then putStrLn "Debugging Haskell is fun!"
else putStrLn "I dont know who you are."
Of course, we neednt put all the putStrLns inside the if statements. We could
instead write:
main = do
putStrLn "Please enter your name:"
name <- getLine
putStrLn
(if name elem ["Simon", "John", "Phil"]
then "Haskell is great!"
else if name == "Koen"
then "Debugging Haskell is fun!"
else "I dont know who you are.")
Using case, we get something like:
171
main = do
putStrLn "Please enter your name:"
name <- getLine
case name of
"Simon" -> putStrLn "Haskell is great!"
"John" -> putStrLn "Haskell is great!"
"Phil" -> putStrLn "Haskell is great!"
"Koen" -> putStrLn "Debugging Haskell is fun!"
_
-> putStrLn "I dont know who you are."
Which, in this case, is actually not much cleaner.
Solution 5.2
The code might look something like:
module DoFile where
import IO
main = do
putStrLn "Do you want to [read] a file, ...?"
cmd <- getLine
case cmd of
"quit" -> return ()
"read" -> do doRead; main
"write" -> do doWrite; main
_
-> do putStrLn
("I dont understand the command "
++ cmd ++ ".")
main
doRead = do
putStrLn "Enter a file name to read:"
fn <- getLine
bracket (openFile fn ReadMode) hClose
(\h -> do txt <- hGetContents h
putStrLn txt)
doWrite = do
putStrLn "Enter a file name to write:"
fn <- getLine
bracket (openFile fn WriteMode) hClose
(\h -> do putStrLn
"Enter text (...):"
172
writeLoop h)
writeLoop h = do
l <- getLine
if l == "."
then return ()
else do hPutStrLn h l
writeLoop h
The only interesting things here are the calls to bracket, which ensure the that the
program lives on, regardless of whether theres a failure or not; and the writeLoop
function. Note that we need to pass the handle returned by openFile (through
bracket to this function, so it knows where to write the input to).
Solution 7.1
Function func3 cannot be converted into point-free style. The others look something
like:
func1 x = map (*x)
func2 f g = filter f . map g
func4 = map (+2) . filter (elem [1..10]) . (5:)
func5 = foldr (uncurry $ flip f) 0
You might have been tempted to try to write func2 as filter f . map,
trying to eta-reduce off the g. In this case, this isnt possible. This is because the
function composition operator (.) has type (b c) (a b) (a c). In this
case, were trying to use map as the second argument. But map takes two arguments,
while (.) expects a function which takes only one.
Solution 7.2
We can start out with a recursive definition:
and [] = True
and (x:xs) = x && and xs
From here, we can clearly rewrite this as:
and = foldr (&&) True
Solution 7.3
We can write this recursively as:
173
concatMap f [] = []
concatMap f (x:xs) = f x ++ concatMap f xs
This hints that we can write this as:
concatMap f = foldr (\a b -> f a ++ b) []
Now, we can do point elimination to get:
==>
==>
==>
==>
foldr
foldr
foldr
foldr
foldr
(\a b -> f a ++ b) []
(\a b -> (++) (f a) b) []
(\a -> (++) (f a)) []
(\a -> ((++) . f) a) []
((++) . f) []
Solution 9.1
The first law is: return a >>= f f a. In the case of Maybe, we get:
==>
==>
==>
return a >>= f
Just a
>>= \x -> f x
(\x -> f x) a
f a
f >>= return
f >>= \x -> return x
f >>= \x -> Just x
At this point, there are two cases depending on whether f is Nothing or not. In
the first case, we get:
==>
==>
==>
174
And the second law is shown. The third law states: f >>= (\x -> g x >>=
h) (f >>= g) >>= h.
If f is Nothing, then the left-hand-side clearly reduces to Nothing. The righthand-side reduces to Nothing >>= h which in turn reduces to Nothing, so they
are the same.
Suppose f is Just a. Then the LHS reduces to g a >>= h and the RHS reduces to (Just a >>= \x -> g x) >>= h which in turn reduces to g a >>=
h, so these two are the same.
Solution 9.2
The idea is that we wish to use the Left constructor to represent errors on the Right
constructor to represent successes. This leads to an instance declaration like:
instance Monad (Either String) where
return x
= Right x
Left s >>= _ = Left s
Right x >>= f = f x
fail s
= Left s
If we try to use this monad to do search, we get:
Monads> searchAll gr 0 3 :: Either String [Int]
Right [0,1,3]
Monads> searchAll gr 3 0 :: Either String [Int]
Left "no path"
which is exactly what we want.
Solution 9.3
The order to mplus essentially determins the search order. When the recursive call to
searchAll2 comes first, we are doing depth-first search. When the recursive call to
search comes first, we are doing breadth-first search. Thus, using the list monad,
we expect the solutions to come in the other order:
MPlus> searchAll3 gr 0 3 :: [[Int]]
[[0,2,3],[0,1,3]]
Just as we expected.
Solution 9.4
This is a very difficult problem; if you found that you were stuck immediately, please
just read as much of this solution as you need to try it yourself.
First, we need to define a list transformer monad. This looks like:
175
176
Solution 9.5
This exercise is actually simpler than the previous one. All we need to do is incorporate
the calls to putT and getT into searchAll6 and add an extra lift to the IO calls.
This extra lift is required because now were stacking two transformers on top of IO
instead of just one.
searchAll7 g@(Graph vl el) src dst
| src == dst = do
lift $ lift $ putStrLn $
"Exploring " ++ show src ++ " -> " ++ show dst
visited <- getT
177
putT (src:visited)
return [src]
| otherwise = do
lift $ lift $ putStrLn $
"Exploring " ++ show src ++ " -> " ++ show dst
visited <- getT
putT (src:visited)
if src elem visited
then mzero
else search el
where
search [] = mzero
search ((u,v,_):es)
| src == u =
(do path <- searchAll7 g v dst
return (u:path)) mplus
search es
| otherwise = search es
The type of this has grown significantly. After applying the graph and two ints, this
has type Monad (t IO), MonadTrans t, MonadPlus (StateT [Int]
(t IO)) => StateT [Int] (t IO) [Int].
Essentially this means that weve got something thats a state transformer wrapped
on top of some other arbitrary transformer (t) which itself sits on top of IO. In our
case, t is going to be ListT. Thus, we run this beast by saying:
MTrans> unListT (evalStateT (searchAll7 gr4 0 3) [])
Exploring 0 -> 3
Exploring 1 -> 3
Exploring 3 -> 3
Exploring 0 -> 3
Exploring 2 -> 3
Exploring 3 -> 3
MTrans> it
[[0,1,3],[0,2,3]]
And it works, even on gr4.
Solution 9.6
First we write a function spaces which will parse out whitespaces:
spaces :: Parser ()
spaces = many (matchChar isSpace) >> return ()
Now, using this, we simply sprinkle calls to spaces through intList to get
intListSpace:
178
\n\n ,5\n]"
\n\n ,a\n]"
Solution 9.7
We do this by replacing the state functions with push and pop functions as follows:
parseValueLet2 :: CharParser (FiniteMap Char [Int]) Int
parseValueLet2 = choice
[ int
, do string "let "
c <- letter
char =
e <- parseValueLet2
string " in "
pushBinding c e
v <- parseValueLet2
popBinding c
return v
, do c <- letter
fm <- getState
case lookupFM fm c of
Nothing
-> unexpected ("variable " ++
show c ++
" unbound")
Just (i:_) -> return i
, between (char () (char )) $ do
e1 <- parseValueLet2
179
op <- oneOf "+*"
e2 <- parseValueLet2
case op of
+ -> return (e1 + e2)
* -> return (e1 * e2)
]
where
pushBinding c v = do
fm <- getState
case lookupFM fm c of
Nothing -> setState (addToFM fm c [v])
Just l -> setState (addToFM fm c (v:l))
popBinding c = do
fm <- getState
case lookupFM fm c of
Just [_]
-> setState (delFromFM fm c)
Just (_:l) -> setState (addToFM fm c l)
The primary difference here is that instead of calling updateState, we use two
local functions, pushBinding and popBinding. The pushBinding function
takes a variable name and a value and adds the value onto the head of the list pointed to
in the state FiniteMap. The popBinding function looks at the value and if there is
only one element on the stack, it completely removes the stack from the FiniteMap;
otherwise it just removes the first element. This means that if something is in the
FiniteMap, the stack is never empty.
This enables us to modify only slightly the usage case; this time, we simply take
the top element off the stack when we need to inspect the value of a variable.
We can test that this works:
ParsecI> runParser parseValueLet2 emptyFM "stdin"
"((let x=2 in 3+4)*x)"
Left "stdin" (line 1, column 20):
unexpected variable x unbound
Index
(), 87
, 32
+, 32
++, 38
, 32
, 52
., 50
.., 140
/, 32
//, 144
:, 36
::, 65
==, 65
[], 36
, 32
E30F , 72
, 72
{}, 52
, 122
as, 111
derive, 134
do, 57
hiding, 111
import, 110
qualified, 110
brackets, 36
buffering, 58
comments, 5253
common sub-expression, 117
Compilers, see GHC,NHC,Interpreters
concatenate, 38
cons, 36
constructors, 81
continuation passing style, 8790
CPS, see continuation passing style
destructive update, 31
do notation, 173176
do notation, 57
drop, 140
Editors, 2728
Emacs, 27
elems, 144
Enum, 131132
enumerated types, 84
enumFromThenTo, 141
enumFromTo, 141
Eq, 127129
equality, 65
equals, 70
evaluation order, 32
exceptions, 182
exports, 107110
expressions, 32
extensions
enabling
in GHC, 25
in Hugs, 23
accumArray, 143
actions, 95100
arithmetic, 3234
array, 143
arrays, 143145
assocs, 144
Bird scripts, 113
boolean, 66
bounds, 144
bracket, 101
fallthrough, 123
false, see boolean
180
INDEX
files, 4244
filter, 38, 138
FiniteMap, 145147
foldl, 38, 139
foldr, 38, 138
fst, 35
functional, 3
functions, 4452
anonymous, see lambda
as arguments, 7880
associative, 40
composition, 50
type, 7180
getChar, 101
getContents, 101
getLine, 58, 101
GHC, 19, 2326
guards, 125127
Haskell 98, 8
Haskell Bookshelf, 7
hClose, 101
head, 37, 74
hGetChar, 101
hGetContents, 101
hGetLin, 101
hIsEOF, 101
hPutChar, 101
hPutStr, 101
hPutStrLn, 101
Hugs, 1923
ICFP, 9
immutable, 31
imports, 110111
indices, 144
induction, 54
infix, 75, 115116
input, see IO
instance, 67
interactive programs, 5663
Interpreters, see GHC,Hugs
IO, 91106
library, 100103
isLower, 38
181
lambda, 72
lambda calculus, 72
LaTeX scripts, 114
layout, 147
lazy, 2, 29
length, 37, 79
let
and do, 175
list
comprehensions, 143
listArray, 143
lists, 3542, 137143
comprehensions, 140
cons, 36
empty, 36
literate style, 112114
local bindings, 117
loop, 53
map, 38, 138
maps, 145147
Maybe, 8587
modules, 107114
hierarchical, 111112
monads, 94
and do, 173176
definition of, 170173
laws, 171
plus, 181
state, 176181
transformer, 181
mutable, see immutable
named fields, 135137
NHC, 19, 2627
null, 74
Num, 132
numeric types, 70
openFile, 101
operator precedence, 33
Ord, 130131
output, see IO
pairs, see tuples
parentheses, 33
INDEX
182
pattern matching, 82, 119125
pure, 2
putChar, 101
putStr, 101
putStrLn, 44, 101
random numbers, 59
randomRIO, 59
Read, 132133
read, 38
readFile, 101
recursion, 5356
referential tranparency, 31
robot, 9
sections, 115116
shadowing, 117
Show, 129130
show, 38, 71
snd, 35
sqrt, 32
standard, 8
state, 2
strict, 2, 152155
strings, 3738
converting from/to, 38
tail, 37, 74
take, 140
toUpper, 38
true, see boolean
tuples, 3435
type, 64, 168
checking, 64
classes, 6971, 155162
instances, 127134, 162165
datatypes, 8087, 134137, 152
155
strict, 152155
default, 168
errors, 66
explicit declarations, 7778
hierarchy, 168
higher-order, 7375
inference, 64
IO, 7577
kinds, 165168
newtype, 149152
polymorphic, 6769
signatures, 77
synonyms, 101, 148149
Unit, 87
unzip, 139
user input, see interactive programs
wildcard, 122
writeFile, 101
zip, 139